added space app files

README.md

@@ -1,13 +1,2 @@
----
-title: Yolov3
-emoji: 📊
-colorFrom: yellow
-colorTo: indigo
-sdk: gradio
-sdk_version: 3.40.1
-app_file: app.py
-pinned: false
-license: mit
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# yolov3
+S13 ERA V1

app.py

@@ -0,0 +1,292 @@
+import gradio as gr
+from typing import List
+import cv2
+import torch
+from torchvision import transforms
+import numpy as np
+from PIL import Image
+from pytorch_grad_cam import GradCAM
+from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
+from pytorch_grad_cam.utils.image import show_cam_on_image
+import io
+from models import YoloV3Lightning
+from utils import load_model_from_checkpoint
+import utils
+import config as cfg
+import matplotlib.pyplot as plt
+import matplotlib.patches as patches
+from dataset import YOLODataset
+from torch.utils.data import Dataset, DataLoader
+from grad_cam import YoloGradCAM
+
+device = torch.device('cpu')
+dataset_mean, dataset_std = (0.4914, 0.4822, 0.4465), \
+                            (0.2470, 0.2435, 0.2616)
+model = YoloV3Lightning.YOLOv3LightningModel(num_classes=cfg.NUM_CLASSES, anchors=cfg.ANCHORS, S=cfg.S)
+ckpt_file = 'ckpt_light.pth'
+checkpoint = load_model_from_checkpoint(device, file_name=ckpt_file)
+model.load_state_dict(checkpoint['model'], strict=False)
+
+model.eval()
+
+scaled_anchors = (
+        torch.tensor(cfg.ANCHORS)
+        * torch.tensor(cfg.S).unsqueeze(1).unsqueeze(1).repeat(1, 3, 2)
+).to(model.device)
+
+cam = YoloGradCAM(model=model, target_layers=[model.layers[-2]], scaled_anchors=scaled_anchors, use_cuda=False)
+'''cfg.IMG_DIR = cfg.DATASET + "/images/"
+cfg.LABEL_DIR = cfg.DATASET + "/labels/"
+eval_dataset = YOLODataset(
+    cfg.DATASET + "/25examples.csv",
+    transform=cfg.test_transforms,
+    S=[cfg.IMAGE_SIZE // 32, cfg.IMAGE_SIZE // 16, cfg.IMAGE_SIZE // 8],
+    img_dir=cfg.IMG_DIR,
+    label_dir=cfg.LABEL_DIR,
+    anchors=cfg.ANCHORS,
+    mosaic=False
+)
+eval_loader = DataLoader(
+    dataset=eval_dataset,
+    batch_size=cfg.BATCH_SIZE,
+    num_workers=cfg.NUM_WORKERS,
+    pin_memory=cfg.PIN_MEMORY,
+    shuffle=True,
+    drop_last=False,
+)
+
+scaled_anchors = (
+        torch.tensor(cfg.ANCHORS)
+        * torch.tensor(cfg.S).unsqueeze(1).unsqueeze(1).repeat(1, 3, 2)
+)
+scaled_anchors = scaled_anchors.to(cfg.DEVICE)
+
+utils.plot_examples(model, eval_loader, 0.5, 0.6, scaled_anchors)'''
+
+sample_images = [
+    ['images/000001.jpg'],
+    ['images/000002.jpg'],
+    ['images/000003.jpg'],
+    ['images/000004.jpg'],
+    ['images/000005.jpg'],
+    ['images/000006.jpg'],
+    ['images/000007.jpg'],
+    ['images/000008.jpg'],
+    ['images/000009.jpg'],
+    ['images/000010.jpg'],
+    ['images/000011.jpg'],
+    ['images/000012.jpg'],
+    ['images/000013.jpg'],
+    ['images/000014.jpg'],
+    ['images/000015.jpg'],
+    ['images/000016.jpg'],
+    ['images/000017.jpg'],
+    ['images/000018.jpg'],
+    ['images/000019.jpg'],
+    ['images/000020.jpg'],
+    ['images/000021.jpg'],
+    ['images/000022.jpg'],
+    ['images/000023.jpg'],
+    ['images/000024.jpg'],
+    ['images/000025.jpg']
+]
+
+with gr.Blocks() as app:
+    with gr.Row():
+
+        gr.Markdown(
+            """
+            # YoloV3 App!
+            ## Model is trained on PASCAL-VOC data to predict following classes -
+            """)
+
+    with gr.Row():
+        gr.HTML(
+            """
+            <table>
+                <tr>
+                    <th>aeroplane</th>
+                    <th>bicycle</th>
+                    <th>bird</th>
+                    <th>boat</th>
+                    <th>bottle</th>
+                    <th>bus</th>
+                    <th>car</th>
+                    <th>cat</th>
+                </tr>
+                <tr>
+                    <th>chair</th>
+                    <th>cow</th>
+                    <th>diningtable</th>
+                    <th>dog</th>
+                    <th>horse</th>
+                    <th>motorbike</th>
+                    <th>person</th>
+                    <th>pottedplant</th>
+                </tr>
+                <tr>
+                    <th>sheep</th>
+                    <th>sofa</th>
+                    <th>train</th>
+                    <th>tvmonitor</th>
+                </tr>
+
+            </table>
+            <p>
+                <a href='https://github.com/piygr/yolov3/blob/main/models/YoloV3Lightning.py'>Click to see the model architecture / code </a>
+
+            </p>
+            """
+        )
+    with gr.Row(visible=True) as top_pred_cls_col:
+        with gr.Column():
+            example_images = gr.Gallery(allow_preview=False, label='Select image ', info='',
+                                        value=[img[0] for img in sample_images], columns=3, rows=2)
+
+        with gr.Column():
+            top_pred_image = gr.Image(label='Upload Image or Select from the gallery')
+
+            with gr.Row():
+                top_class_btn = gr.Button("Submit", variant='primary')
+                tc_clear_btn = gr.ClearButton()
+
+    with gr.Row():
+        if_show_grad_cam = gr.Checkbox(value=True, label='Show Class Activation Map (What the model sees)?')
+
+    # with gr.Row(visible=True) as top_class_output:
+    with gr.Row(visible=True)  as top_class_output:
+        top_class_output_img = gr.Image(interactive=False, label='Prediction Output')
+    with gr.Row(visible=True)  as top_class_output:
+        grad_cam_out = gr.Image(interactive=False, visible=True, label='CAM Outcome')
+
+
+    def show_cam_output(input):
+        return {
+            grad_cam_out: gr.update(visible=input)
+        }
+
+
+    if_show_grad_cam.change(
+        show_cam_output,
+        if_show_grad_cam,
+        grad_cam_out
+    )
+
+
+    def clear_data():
+        return {
+            top_pred_image: None,
+            top_class_output_img: None
+        }
+
+
+    tc_clear_btn.click(clear_data, None, [top_pred_image, top_class_output_img])
+
+
+    def on_select(evt: gr.SelectData):
+        return {
+            top_pred_image: sample_images[evt.index][0]
+        }
+
+
+    example_images.select(on_select, None, top_pred_image)
+
+
+    def plot_image(image, boxes):
+        """Plots predicted bounding boxes on the image"""
+        cmap = plt.get_cmap("tab20b")
+        class_labels = cfg.PASCAL_CLASSES
+        colors = [cmap(i) for i in np.linspace(0, 1, len(class_labels))]
+        im = np.array(image)
+        height, width, _ = im.shape
+
+        # Create figure and axes
+        fig, ax = plt.subplots(1)
+        # Display the image
+        ax.imshow(im)
+
+        # box[0] is x midpoint, box[2] is width
+        # box[1] is y midpoint, box[3] is height
+
+        # Create a Rectangle patch
+        for box in boxes:
+            assert len(box) == 6, "box should contain class pred, confidence, x, y, width, height"
+            class_pred = box[0]
+            box = box[2:]
+            upper_left_x = box[0] - box[2] / 2
+            upper_left_y = box[1] - box[3] / 2
+            rect = patches.Rectangle(
+                (upper_left_x * width, upper_left_y * height),
+                box[2] * width,
+                box[3] * height,
+                linewidth=2,
+                edgecolor=colors[int(class_pred)],
+                facecolor="none",
+            )
+            # Add the patch to the Axes
+            ax.add_patch(rect)
+            plt.text(
+                upper_left_x * width,
+                upper_left_y * height,
+                s=class_labels[int(class_pred)],
+                color="white",
+                verticalalignment="top",
+                bbox={"color": colors[int(class_pred)], "pad": 0},
+            )
+
+        plt.savefig('output.png')
+        x = plt.show()
+
+
+    def predict(image: np.ndarray, iou_thresh: float = 0.5, thresh: float = 0.4, show_cam: bool = False,
+                transparency: float = 0.5) -> List[np.ndarray]:
+        with torch.no_grad():
+            transformed_image = cfg.grad_cam_transforms(image=image)["image"].unsqueeze(0)
+            output = model(transformed_image)
+
+            bboxes = [[] for _ in range(1)]
+            for i in range(3):
+                batch_size, A, S, _, _ = output[i].shape
+                anchor = scaled_anchors[i]
+                boxes_scale_i = utils.cells_to_bboxes(
+                    output[i], anchor, S=S, is_preds=True
+                )
+                for idx, (box) in enumerate(boxes_scale_i):
+                    bboxes[idx] += box
+
+        nms_boxes = utils.non_max_suppression(
+            bboxes[0], iou_threshold=iou_thresh, threshold=thresh, box_format="midpoint",
+        )
+
+        plot_image(image, nms_boxes)
+        plotted_img = 'output.png'
+        if not show_cam:
+            return [plotted_img, None]
+
+        grayscale_cam = cam(transformed_image)[0, :, :]
+        img = cv2.resize(image, (416, 416))
+        img = np.float32(img) / 255
+        cam_image = show_cam_on_image(img, grayscale_cam, use_rgb=True, image_weight=transparency)
+        return [plotted_img, cam_image]
+
+
+    def top_class_img_upload(input_img, if_cam):
+        if input_img is not None:
+            imgs = predict(input_img, show_cam=if_cam)
+
+            return {
+                top_class_output_img: imgs[0],
+                grad_cam_out: imgs[1]
+            }
+
+
+    top_class_btn.click(
+        top_class_img_upload,
+        [top_pred_image, if_show_grad_cam],
+        [top_class_output_img, grad_cam_out]
+    )
+
+'''
+Launch the app
+'''
+app.launch()

config.py

@@ -0,0 +1,113 @@
+import albumentations as A
+import cv2
+import torch
+
+from albumentations.pytorch import ToTensorV2
+
+SAMPLE_DATASET = '../../input/d/piygro/sample-pascal/PASCAL_VOC'
+DATASET = 'PASCAL_VOC'
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+# seed_everything()  # If you want deterministic behavior
+NUM_WORKERS = 2
+BATCH_SIZE = 16
+IMAGE_SIZE = 416
+NUM_CLASSES = 20
+LEARNING_RATE = 1e-3
+WEIGHT_DECAY = 1e-4
+NUM_EPOCHS = 40
+CONF_THRESHOLD = 0.05
+MAP_IOU_THRESH = 0.5
+NMS_IOU_THRESH = 0.45
+S = [IMAGE_SIZE // 32, IMAGE_SIZE // 16, IMAGE_SIZE // 8]
+PIN_MEMORY = True
+LOAD_MODEL = False
+SAVE_MODEL = True
+CHECKPOINT_FILE = "checkpoint.ckpt.tar"
+IMG_DIR = DATASET + "/images/"
+LABEL_DIR = DATASET + "/labels/"
+
+ANCHORS = [
+    [(0.28, 0.22), (0.38, 0.48), (0.9, 0.78)],
+    [(0.07, 0.15), (0.15, 0.11), (0.14, 0.29)],
+    [(0.02, 0.03), (0.04, 0.07), (0.08, 0.06)],
+]  # Note these have been rescaled to be between [0, 1]
+
+means = [0.485, 0.456, 0.406]
+
+scale = 1.1
+train_transforms = A.Compose(
+    [
+        A.LongestMaxSize(max_size=int(IMAGE_SIZE * scale)),
+        A.PadIfNeeded(
+            min_height=int(IMAGE_SIZE * scale),
+            min_width=int(IMAGE_SIZE * scale),
+            border_mode=cv2.BORDER_CONSTANT,
+        ),
+        A.Rotate(limit = 10, interpolation=1, border_mode=4),
+        A.RandomCrop(width=IMAGE_SIZE, height=IMAGE_SIZE),
+        A.ColorJitter(brightness=0.6, contrast=0.6, saturation=0.6, hue=0.6, p=0.4),
+        A.OneOf(
+            [
+                A.ShiftScaleRotate(
+                    rotate_limit=20, p=0.5, border_mode=cv2.BORDER_CONSTANT
+                ),
+                # A.Affine(shear=15, p=0.5, mode="constant"),
+            ],
+            p=1.0,
+        ),
+        A.HorizontalFlip(p=0.5),
+        A.Blur(p=0.1),
+        A.CLAHE(p=0.1),
+        A.Posterize(p=0.1),
+        A.ToGray(p=0.1),
+        A.ChannelShuffle(p=0.05),
+        A.Normalize(mean=[0, 0, 0], std=[1, 1, 1], max_pixel_value=255,),
+        ToTensorV2(),
+    ],
+    bbox_params=A.BboxParams(format="yolo", min_visibility=0.4, label_fields=[],),
+)
+test_transforms = A.Compose(
+    [
+        A.LongestMaxSize(max_size=IMAGE_SIZE),
+        A.PadIfNeeded(
+            min_height=IMAGE_SIZE, min_width=IMAGE_SIZE, border_mode=cv2.BORDER_CONSTANT
+        ),
+        A.Normalize(mean=[0, 0, 0], std=[1, 1, 1], max_pixel_value=255,),
+        ToTensorV2(),
+    ],
+    bbox_params=A.BboxParams(format="yolo", min_visibility=0.4, label_fields=[]),
+)
+
+grad_cam_transforms = A.Compose(
+    [
+        A.LongestMaxSize(max_size=IMAGE_SIZE),
+        A.PadIfNeeded(
+            min_height=IMAGE_SIZE, min_width=IMAGE_SIZE, border_mode=cv2.BORDER_CONSTANT
+        ),
+        A.Normalize(mean=[0, 0, 0], std=[1, 1, 1], max_pixel_value=255,),
+        ToTensorV2(),
+    ],
+)
+
+PASCAL_CLASSES = [
+    "aeroplane",
+    "bicycle",
+    "bird",
+    "boat",
+    "bottle",
+    "bus",
+    "car",
+    "cat",
+    "chair",
+    "cow",
+    "diningtable",
+    "dog",
+    "horse",
+    "motorbike",
+    "person",
+    "pottedplant",
+    "sheep",
+    "sofa",
+    "train",
+    "tvmonitor"
+]

dataset.py

@@ -0,0 +1,184 @@
+"""
+Creates a Pytorch dataset to load the Pascal VOC & MS COCO datasets
+"""
+import config as cfg
+import numpy as np
+import os
+import pandas as pd
+import torch
+from utils import xywhn2xyxy, xyxy2xywhn
+import random
+
+from PIL import Image, ImageFile
+from torch.utils.data import Dataset, DataLoader
+from utils import (
+    cells_to_bboxes,
+    iou_width_height as iou,
+    non_max_suppression as nms,
+    plot_image
+)
+
+ImageFile.LOAD_TRUNCATED_IMAGES = True
+
+
+class YOLODataset(Dataset):
+    def __init__(
+            self,
+            csv_file,
+            img_dir,
+            label_dir,
+            anchors,
+            image_size=416,
+            S=[13, 26, 52],
+            C=20,
+            transform=None,
+            mosaic=True
+    ):
+        self.annotations = pd.read_csv(csv_file)
+        self.img_dir = img_dir
+        self.label_dir = label_dir
+        self.image_size = image_size
+        self.mosaic_border = [image_size // 2, image_size // 2]
+        self.transform = transform
+        self.S = S
+        self.anchors = torch.tensor(anchors[0] + anchors[1] + anchors[2])  # for all 3 scales
+        self.num_anchors = self.anchors.shape[0]
+        self.num_anchors_per_scale = self.num_anchors // 3
+        self.C = C
+        self.ignore_iou_thresh = 0.5
+        self.mosaic = mosaic
+
+    def __len__(self):
+        return len(self.annotations)
+
+    def load_mosaic(self, index):
+        # YOLOv5 4-mosaic loader. Loads 1 image + 3 random images into a 4-image mosaic
+        labels4 = []
+        s = self.image_size
+        yc, xc = (int(random.uniform(x, 2*s - x)) for x in self.mosaic_border)  # mosaic center x, y
+        indices = [index] + random.choices(range(len(self)), k=3)  # 3 additional image indices
+        random.shuffle(indices)
+        for i, index in enumerate(indices):
+            # Load image
+            label_path = os.path.join(self.label_dir, self.annotations.iloc[index, 1])
+            bboxes = np.roll(np.loadtxt(fname=label_path, delimiter=" ", ndmin=2), 4, axis=1).tolist()
+            img_path = os.path.join(self.img_dir, self.annotations.iloc[index, 0])
+            img = np.array(Image.open(img_path).convert("RGB"))
+
+            h, w = img.shape[0], img.shape[1]
+            labels = np.array(bboxes)
+
+            # place img in img4
+            if i == 0:  # top left
+                img4 = np.full((s * 2, s * 2, img.shape[2]), 114, dtype=np.uint8)  # base image with 4 tiles
+                x1a, y1a, x2a, y2a = max(xc - w, 0), max(yc - h, 0), xc, yc  # xmin, ymin, xmax, ymax (large image)
+                x1b, y1b, x2b, y2b = w - (x2a - x1a), h - (y2a - y1a), w, h  # xmin, ymin, xmax, ymax (small image)
+            elif i == 1:  # top right
+                x1a, y1a, x2a, y2a = xc, max(yc - h, 0), min(xc + w, s * 2), yc
+                x1b, y1b, x2b, y2b = 0, h - (y2a - y1a), min(w, x2a - x1a), h
+            elif i == 2:  # bottom left
+                x1a, y1a, x2a, y2a = max(xc - w, 0), yc, xc, min(s * 2, yc + h)
+                x1b, y1b, x2b, y2b = w - (x2a - x1a), 0, w, min(y2a - y1a, h)
+            elif i == 3:  # bottom right
+                x1a, y1a, x2a, y2a = xc, yc, min(xc + w, s * 2), min(s * 2, yc + h)
+                x1b, y1b, x2b, y2b = 0, 0, min(w, x2a - x1a), min(y2a - y1a, h)
+
+            img4[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b]  # img4[ymin:ymax, xmin:xmax]
+            padw = x1a - x1b
+            padh = y1a - y1b
+
+            # Labels
+            if labels.size:
+                labels[:, :-1] = xywhn2xyxy(labels[:, :-1], w, h, padw, padh)  # normalized xywh to pixel xyxy format
+            labels4.append(labels)
+
+        # Concat/clip labels
+        labels4 = np.concatenate(labels4, 0)
+        for x in (labels4[:, :-1],):
+            np.clip(x, 0, 2 * s, out=x)  # clip when using random_perspective()
+        # img4, labels4 = replicate(img4, labels4)  # replicate
+        labels4[:, :-1] = xyxy2xywhn(labels4[:, :-1], 2 * s, 2 * s)
+        labels4[:, :-1] = np.clip(labels4[:, :-1], 0, 1)
+        labels4 = labels4[labels4[:, 2] > 0]
+        labels4 = labels4[labels4[:, 3] > 0]
+        return img4, labels4
+
+    def __getitem__(self, index):
+
+        if self.mosaic and random.random() <= 0.75:
+            image, bboxes = self.load_mosaic(index)
+        else:
+            label_path = os.path.join(self.label_dir, self.annotations.iloc[index, 1])
+            bboxes = np.roll(np.loadtxt(fname=label_path, delimiter=" ", ndmin=2), 4, axis=1).tolist()
+            img_path = os.path.join(self.img_dir, self.annotations.iloc[index, 0])
+            image = np.array(Image.open(img_path).convert("RGB"))
+
+        if self.transform:
+            augmentations = self.transform(image=image, bboxes=bboxes)
+            image = augmentations["image"]
+            bboxes = augmentations["bboxes"]
+
+        # Below assumes 3 scale predictions (as paper) and same num of anchors per scale
+        targets = [torch.zeros((self.num_anchors // 3, S, S, 6)) for S in self.S]
+        for box in bboxes:
+            iou_anchors = iou(torch.tensor(box[2:4]), self.anchors)
+            anchor_indices = iou_anchors.argsort(descending=True, dim=0)
+            x, y, width, height, class_label = box
+            has_anchor = [False] * 3  # each scale should have one anchor
+            for anchor_idx in anchor_indices:
+                scale_idx = anchor_idx // self.num_anchors_per_scale
+                anchor_on_scale = anchor_idx % self.num_anchors_per_scale
+                S = self.S[scale_idx]
+                i, j = int(S * y), int(S * x)  # which cell
+                anchor_taken = targets[scale_idx][anchor_on_scale, i, j, 0]
+                if not anchor_taken and not has_anchor[scale_idx]:
+                    targets[scale_idx][anchor_on_scale, i, j, 0] = 1
+                    x_cell, y_cell = S * x - j, S * y - i  # both between [0,1]
+                    width_cell, height_cell = (
+                        width * S,
+                        height * S,
+                    )  # can be greater than 1 since it's relative to cell
+                    box_coordinates = torch.tensor(
+                        [x_cell, y_cell, width_cell, height_cell]
+                    )
+                    targets[scale_idx][anchor_on_scale, i, j, 1:5] = box_coordinates
+                    targets[scale_idx][anchor_on_scale, i, j, 5] = int(class_label)
+                    has_anchor[scale_idx] = True
+
+                elif not anchor_taken and iou_anchors[anchor_idx] > self.ignore_iou_thresh:
+                    targets[scale_idx][anchor_on_scale, i, j, 0] = -1  # ignore prediction
+
+        return image, tuple(targets)
+
+
+def validate_dataset():
+    anchors = cfg.ANCHORS
+
+    transform = cfg.test_transforms
+
+    dataset = YOLODataset(
+        cfg.SAMPLE_DATASET + "/25examples.csv",
+        cfg.SAMPLE_DATASET + "/images/",
+        cfg.SAMPLE_DATASET + "/labels/",
+        S=[13, 26, 52],
+        anchors=anchors,
+        transform=transform,
+    )
+    S = [13, 26, 52]
+    scaled_anchors = torch.tensor(anchors) / (
+            1 / torch.tensor(S).unsqueeze(1).unsqueeze(1).repeat(1, 3, 2)
+    )
+    loader = DataLoader(dataset=dataset, batch_size=1, shuffle=True)
+    for x, y in loader:
+        boxes = []
+
+        for i in range(y[0].shape[1]):
+            anchor = scaled_anchors[i]
+            print(anchor.shape)
+            print(y[i].shape)
+            boxes += cells_to_bboxes(
+                y[i], is_preds=False, S=y[i].shape[2], anchors=anchor
+            )[0]
+        boxes = nms(boxes, iou_threshold=1, threshold=0.7, box_format="midpoint")
+        print(boxes)
+        plot_image(x[0].permute(1, 2, 0).to("cpu"), boxes)

grad_cam.py

@@ -0,0 +1,83 @@
+from typing import List
+import torch
+import numpy as np
+import utils
+from pytorch_grad_cam.base_cam import BaseCAM
+from pytorch_grad_cam.utils import get_2d_projection
+from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
+
+class YoloGradCAM(BaseCAM):
+    def __init__(self,
+                 model,
+                 target_layers,
+                 scaled_anchors,
+                 use_cuda=False,
+                 reshape_transform=None):
+        super(YoloGradCAM, self).__init__(model,
+                                      target_layers,
+                                      use_cuda,
+                                      reshape_transform,
+                                      uses_gradients=False)
+
+        self.scaled_anchors = scaled_anchors
+
+    def get_cam_image(self,
+                      input_tensor: torch.Tensor,
+                      target_layer: torch.nn.Module,
+                      targets: List[torch.nn.Module],
+                      activations: torch.Tensor,
+                      grads: torch.Tensor,
+                      eigen_smooth: bool = False) -> np.ndarray:
+        return get_2d_projection(activations)
+
+    def forward(self,
+                input_tensor: torch.Tensor,
+                targets: List[torch.nn.Module],
+                eigen_smooth: bool = False) -> np.ndarray:
+
+        if self.cuda:
+            input_tensor = input_tensor.cuda()
+
+        if self.compute_input_gradient:
+            input_tensor = torch.autograd.Variable(input_tensor,
+                                                   requires_grad=True)
+
+        outputs = self.activations_and_grads(input_tensor)
+        if targets is None:
+            bboxes = [[] for _ in range(1)]
+            for i in range(3):
+                batch_size, A, S, _, _ = outputs[i].shape
+                anchor = self.scaled_anchors[i]
+                boxes_scale_i = utils.cells_to_bboxes(
+                    outputs[i], anchor, S=S, is_preds=True
+                )
+                for idx, (box) in enumerate(boxes_scale_i):
+                    bboxes[idx] += box
+
+            nms_boxes = utils.non_max_suppression(
+                bboxes[0], iou_threshold=0.5, threshold=0.4, box_format="midpoint",
+            )
+            # target_categories = np.argmax(outputs.cpu().data.numpy(), axis=-1)
+            target_categories = [box[0] for box in nms_boxes]
+            targets = [ClassifierOutputTarget(
+                category) for category in target_categories]
+
+        if self.uses_gradients:
+            self.model.zero_grad()
+            loss = sum([target(output)
+                        for target, output in zip(targets, outputs)])
+            loss.backward(retain_graph=True)
+
+        # In most of the saliency attribution papers, the saliency is
+        # computed with a single target layer.
+        # Commonly it is the last convolutional layer.
+        # Here we support passing a list with multiple target layers.
+        # It will compute the saliency image for every image,
+        # and then aggregate them (with a default mean aggregation).
+        # This gives you more flexibility in case you just want to
+        # use all conv layers for example, all Batchnorm layers,
+        # or something else.
+        cam_per_layer = self.compute_cam_per_layer(input_tensor,
+                                                   targets,
+                                                   eigen_smooth)
+        return self.aggregate_multi_layers(cam_per_layer)

loss.py

@@ -0,0 +1,149 @@
+"""
+Implementation of Yolo Loss Function similar to the one in Yolov3 paper,
+the difference from what I can tell is I use CrossEntropy for the classes
+instead of BinaryCrossEntropy.
+"""
+import random
+import torch
+import torch.nn as nn
+import pytorch_lightning as pl
+from utils import intersection_over_union
+import config as cfg
+
+
+class YoloLoss(pl.LightningModule):
+    def __init__(self):
+        super().__init__()
+        self.mse = nn.MSELoss()
+        self.bce = nn.BCEWithLogitsLoss()
+        self.entropy = nn.CrossEntropyLoss()
+        self.sigmoid = nn.Sigmoid()
+
+        # Constants signifying how much to pay for each respective part of the loss
+        self.lambda_class = 1
+        self.lambda_noobj = 10
+        self.lambda_obj = 1
+        self.lambda_box = 10
+
+        self.scaled_anchors = (
+            torch.tensor(cfg.ANCHORS)
+            * torch.tensor(cfg.S).unsqueeze(1).unsqueeze(1).repeat(1, 3, 2)
+        )
+
+    def forward(self, predictions_list, target_list, **kwargs):
+
+        anchors_list = kwargs.get('anchors_list', None)
+        if not anchors_list:
+            anchors_list = self.scaled_anchors
+
+        anchors_list = anchors_list.to(cfg.DEVICE)
+
+        box_loss = 0.0
+        object_loss = 0.0
+        no_object_loss = 0.0
+        class_loss = 0.0
+
+        for i in range(3):
+            target = target_list[i]
+            predictions = predictions_list[i]
+            anchors = anchors_list[i]
+
+            # Check where obj and noobj (we ignore if target == -1)
+            obj = target[..., 0] == 1  # in paper this is Iobj_i
+            noobj = target[..., 0] == 0  # in paper this is Inoobj_i
+
+            # ======================= #
+            #   FOR NO OBJECT LOSS    #
+            # ======================= #
+
+            no_object_loss += self.bce(
+                (predictions[..., 0:1][noobj]), (target[..., 0:1][noobj]),
+            )
+
+            # ==================== #
+            #   FOR OBJECT LOSS    #
+            # ==================== #
+
+            anchors = anchors.reshape(1, 3, 1, 1, 2)
+            box_preds = torch.cat([self.sigmoid(predictions[..., 1:3]), torch.exp(predictions[..., 3:5]) * anchors], dim=-1)
+            ious = intersection_over_union(box_preds[obj], target[..., 1:5][obj]).detach()
+            object_loss += self.mse(self.sigmoid(predictions[..., 0:1][obj]), ious * target[..., 0:1][obj])
+
+            # ======================== #
+            #   FOR BOX COORDINATES    #
+            # ======================== #
+
+            predictions[..., 1:3] = self.sigmoid(predictions[..., 1:3])  # x,y coordinates
+            target[..., 3:5] = torch.log(
+                (1e-16 + target[..., 3:5] / anchors)
+            )  # width, height coordinates
+            box_loss += self.mse(predictions[..., 1:5][obj], target[..., 1:5][obj])
+
+            # ================== #
+            #   FOR CLASS LOSS   #
+            # ================== #
+
+            class_loss += self.entropy(
+                (predictions[..., 5:][obj]), (target[..., 5][obj].long()),
+            )
+
+            #print("__________________________________")
+            #print(self.lambda_box * box_loss)
+            #print(self.lambda_obj * object_loss)
+            #print(self.lambda_noobj * no_object_loss)
+            #print(self.lambda_class * class_loss)
+            #print("\n")
+
+        total_loss = (
+            self.lambda_box * box_loss
+            + self.lambda_obj * object_loss
+            + self.lambda_noobj * no_object_loss
+            + self.lambda_class * class_loss
+        )
+
+        if kwargs.get('loss_dict'):
+            return dict(class_loss=self.lambda_class * class_loss,
+                    no_object_loss=self.lambda_noobj * no_object_loss,
+                    object_loss=self.lambda_obj * object_loss,
+                    box_loss=self.lambda_box * box_loss,
+                    total_loss=total_loss
+                    )
+        else:
+            return total_loss
+
+
+    def check_class_accuracy(self, predictions, target, threshold):
+        tot_class_preds, correct_class = 0, 0
+        tot_noobj, correct_noobj = 0, 0
+        tot_obj, correct_obj = 0, 0
+
+        y = target
+        out = predictions
+
+        for i in range(3):
+            obj = y[i][..., 0] == 1  # in paper this is Iobj_i
+            noobj = y[i][..., 0] == 0  # in paper this is Iobj_i
+
+            correct_class += torch.sum(
+                torch.argmax(out[i][..., 5:][obj], dim=-1) == y[i][..., 5][obj]
+            )
+            tot_class_preds += torch.sum(obj)
+
+            obj_preds = torch.sigmoid(out[i][..., 0]) > threshold
+            correct_obj += torch.sum(obj_preds[obj] == y[i][..., 0][obj])
+            tot_obj += torch.sum(obj)
+            correct_noobj += torch.sum(obj_preds[noobj] == y[i][..., 0][noobj])
+            tot_noobj += torch.sum(noobj)
+
+        return dict(
+            correct_class=correct_class,
+            correct_noobj=correct_noobj,
+            correct_obj=correct_obj,
+            total_class_preds=tot_class_preds,
+            total_noobj=tot_noobj,
+            total_obj=tot_obj
+        )
+
+        '''print(f"Class accuracy is: {(correct_class/(tot_class_preds+1e-16))*100:2f}%")
+        print(f"No obj accuracy is: {(correct_noobj/(tot_noobj+1e-16))*100:2f}%")
+        print(f"Obj accuracy is: {(correct_obj/(tot_obj+1e-16))*100:2f}%")'''

main.py

@@ -0,0 +1,86 @@
+from dataset import *
+from models.YoloV3Lightning import *
+import utils
+
+def init(model, basic_sanity_check=True, find_max_lr=True, train=True, **kwargs):
+    if basic_sanity_check:
+        validate_dataset()
+        sanity_check(model)
+        print("Set basic_sanity_check to False to proceed")
+    else:
+        if find_max_lr:
+            optimizer = kwargs.get('optimizer')
+            criterion = kwargs.get('criterion')
+            train_loader = kwargs.get('train_loader')
+            utils.find_lr(model, optimizer, criterion, train_loader)
+            print("Set find_max_lr to False to proceed further")
+        else:
+
+            train_loader = kwargs.get('train_loader')
+            val_loader = kwargs.get('test_loader')
+
+            if train:
+                trainer = pl.Trainer(
+                    precision=16,
+                    max_epochs=cfg.NUM_EPOCHS,
+                    accelerator='gpu'
+                )
+
+                cargs = {}
+                if cfg.LOAD_MODEL:
+                    cargs = dict(ckpt_path=cfg.CHECKPOINT_FILE)
+
+                trainer.fit(model, train_loader, val_loader, **cargs)
+            else:
+                ckpt_file = kwargs.get('ckpt_file')
+                if ckpt_file:
+                    checkpoint = utils.load_model_from_checkpoint(cfg.DEVICE, file_name=ckpt_file)
+                    model.load_state_dict(checkpoint['model'], strict=False)
+
+            #-- Printing samples
+            model.to(cfg.DEVICE)
+            model.eval()
+            cfg.IMG_DIR = cfg.DATASET + "/images/"
+            cfg.LABEL_DIR = cfg.DATASET + "/labels/"
+            eval_dataset = YOLODataset(
+                cfg.DATASET + "/test.csv",
+                transform=cfg.test_transforms,
+                S=[cfg.IMAGE_SIZE // 32, cfg.IMAGE_SIZE // 16, cfg.IMAGE_SIZE // 8],
+                img_dir=cfg.IMG_DIR,
+                label_dir=cfg.LABEL_DIR,
+                anchors=cfg.ANCHORS,
+                mosaic=False
+            )
+            eval_loader = DataLoader(
+                dataset=eval_dataset,
+                batch_size=cfg.BATCH_SIZE,
+                num_workers=cfg.NUM_WORKERS,
+                pin_memory=cfg.PIN_MEMORY,
+                shuffle=True,
+                drop_last=False,
+            )
+
+            scaled_anchors = (
+                    torch.tensor(cfg.ANCHORS)
+                    * torch.tensor(cfg.S).unsqueeze(1).unsqueeze(1).repeat(1, 3, 2)
+            )
+            scaled_anchors = scaled_anchors.to(cfg.DEVICE)
+
+            utils.plot_examples(model, eval_loader, 0.5, 0.6, scaled_anchors)
+
+            # -- Printing MAP
+            pred_boxes, true_boxes = utils.get_evaluation_bboxes(
+                eval_loader,
+                model,
+                iou_threshold=cfg.NMS_IOU_THRESH,
+                anchors=cfg.ANCHORS,
+                threshold=cfg.CONF_THRESHOLD,
+            )
+            mapval = utils.mean_average_precision(
+                pred_boxes,
+                true_boxes,
+                iou_threshold=cfg.MAP_IOU_THRESH,
+                box_format="midpoint",
+                num_classes=cfg.NUM_CLASSES,
+            )
+            print(f"MAP: {mapval.item()}")

models/YoloV3Lightning.py

@@ -0,0 +1,378 @@
+import torch
+import torch.nn as nn
+import pytorch_lightning as pl
+from pytorch_grad_cam import GradCAM
+from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
+from pytorch_grad_cam.utils.image import show_cam_on_image
+import numpy as np
+from torchvision import transforms
+
+import matplotlib.pyplot as plt
+
+from loss import YoloLoss
+import config as cfg
+
+""" 
+Information about architecture config:
+Tuple is structured by (filters, kernel_size, stride) 
+Every conv is a same convolution. 
+List is structured by "B" indicating a residual block followed by the number of repeats
+"S" is for scale prediction block and computing the yolo loss
+"U" is for upsampling the feature map and concatenating with a previous layer
+"""
+config = [
+    (32, 3, 1),
+    (64, 3, 2),
+    ["B", 1],
+    (128, 3, 2),
+    ["B", 2],
+    (256, 3, 2),
+    ["B", 8],
+    (512, 3, 2),
+    ["B", 8],
+    (1024, 3, 2),
+    ["B", 4],  # To this point is Darknet-53
+    (512, 1, 1),
+    (1024, 3, 1),
+    "S",
+    (256, 1, 1),
+    "U",
+    (256, 1, 1),
+    (512, 3, 1),
+    "S",
+    (128, 1, 1),
+    "U",
+    (128, 1, 1),
+    (256, 3, 1),
+    "S",
+]
+
+
+class CNNBlock(nn.Module):
+    def __init__(self, in_channels, out_channels, bn_act=True, **kwargs):
+        super().__init__()
+        self.conv = nn.Conv2d(in_channels, out_channels, bias=not bn_act, **kwargs)
+        self.bn = nn.BatchNorm2d(out_channels)
+        self.leaky = nn.LeakyReLU(0.1)
+        self.use_bn_act = bn_act
+
+    def forward(self, x):
+        if self.use_bn_act:
+            return self.leaky(self.bn(self.conv(x)))
+        else:
+            return self.conv(x)
+
+
+class ResidualBlock(nn.Module):
+    def __init__(self, channels, use_residual=True, num_repeats=1):
+        super().__init__()
+        self.layers = nn.ModuleList()
+        for repeat in range(num_repeats):
+            self.layers += [
+                nn.Sequential(
+                    CNNBlock(channels, channels // 2, kernel_size=1),
+                    CNNBlock(channels // 2, channels, kernel_size=3, padding=1),
+                )
+            ]
+
+        self.use_residual = use_residual
+        self.num_repeats = num_repeats
+
+    def forward(self, x):
+        for layer in self.layers:
+            if self.use_residual:
+                x = x + layer(x)
+            else:
+                x = layer(x)
+
+        return x
+
+
+class ScalePrediction(nn.Module):
+    def __init__(self, in_channels, num_classes):
+        super().__init__()
+        self.pred = nn.Sequential(
+            CNNBlock(in_channels, 2 * in_channels, kernel_size=3, padding=1),
+            CNNBlock(
+                2 * in_channels, (num_classes + 5) * 3, bn_act=False, kernel_size=1
+            ),
+        )
+        self.num_classes = num_classes
+
+    def forward(self, x):
+        return (
+            self.pred(x)
+            .reshape(x.shape[0], 3, self.num_classes + 5, x.shape[2], x.shape[3])
+            .permute(0, 1, 3, 4, 2)
+        )
+
+
+class YOLOv3LightningModel(pl.LightningModule):
+    def __init__(self, in_channels=3, num_classes=20, anchors=None, S=None):
+        super().__init__()
+        self.num_classes = num_classes
+        self.in_channels = in_channels
+        self.layers = self._create_conv_layers()
+        self.anchor_list = (
+                torch.tensor(anchors)
+                * torch.tensor(S).unsqueeze(1).unsqueeze(1).repeat(1, 3, 2)
+        )
+
+        self.criterion = YoloLoss()
+
+        self.metric = dict(
+            total_train_steps=0,
+            epoch_train_loss=[],
+            epoch_train_acc=[],
+            epoch_train_steps=0,
+            total_val_steps=0,
+            epoch_val_loss=[],
+            epoch_val_acc=[],
+            epoch_val_steps=0,
+            train_loss=[],
+            val_loss=[],
+            train_acc=[],
+            val_acc=[]
+        )
+
+    def forward(self, x):
+        outputs = []  # for each scale
+        route_connections = []
+        for layer in self.layers:
+            if isinstance(layer, ScalePrediction):
+                outputs.append(layer(x))
+                continue
+
+            x = layer(x)
+
+            if isinstance(layer, ResidualBlock) and layer.num_repeats == 8:
+                route_connections.append(x)
+
+            elif isinstance(layer, nn.Upsample):
+                x = torch.cat([x, route_connections[-1]], dim=1)
+                route_connections.pop()
+
+        return outputs
+
+    def _create_conv_layers(self):
+        layers = nn.ModuleList()
+        in_channels = self.in_channels
+
+        for module in config:
+            if isinstance(module, tuple):
+                out_channels, kernel_size, stride = module
+                layers.append(
+                    CNNBlock(
+                        in_channels,
+                        out_channels,
+                        kernel_size=kernel_size,
+                        stride=stride,
+                        padding=1 if kernel_size == 3 else 0,
+                    )
+                )
+                in_channels = out_channels
+
+            elif isinstance(module, list):
+                num_repeats = module[1]
+                layers.append(ResidualBlock(in_channels, num_repeats=num_repeats,))
+
+            elif isinstance(module, str):
+                if module == "S":
+                    layers += [
+                        ResidualBlock(in_channels, use_residual=False, num_repeats=1),
+                        CNNBlock(in_channels, in_channels // 2, kernel_size=1),
+                        ScalePrediction(in_channels // 2, num_classes=self.num_classes),
+                    ]
+                    in_channels = in_channels // 2
+
+                elif module == "U":
+                    layers.append(nn.Upsample(scale_factor=2),)
+                    in_channels = in_channels * 3
+
+        return layers
+
+
+    def get_layer(self, idx):
+        if idx < len(self.layers) and idx >= 0:
+            return self.layers[idx]
+
+
+
+    def training_step(self, train_batch, batch_idx):
+        x, target = train_batch
+        output = self.forward(x)
+        loss = self.criterion(output, target, loss_dict=True, anchor_list=self.anchor_list)
+        acc = self.criterion.check_class_accuracy(output, target, cfg.CONF_THRESHOLD)
+
+        self.metric['total_train_steps'] += 1
+        self.metric['epoch_train_steps'] += 1
+        self.metric['epoch_train_loss'].append(loss)
+        self.metric['epoch_train_acc'].append(acc)
+
+        self.log_dict({'train_loss': loss['total_loss']})
+
+        return loss['total_loss']
+
+
+    def validation_step(self, val_batch, batch_idx):
+        x, target = val_batch
+        output = self.forward(x)
+        loss = self.criterion(output, target, loss_dict=True, anchor_list=self.anchor_list)
+        acc = self.criterion.check_class_accuracy(output, target, cfg.CONF_THRESHOLD)
+
+        self.metric['total_val_steps'] += 1
+        self.metric['epoch_val_steps'] += 1
+        self.metric['epoch_val_loss'].append(loss)
+        self.metric['epoch_val_acc'].append(acc)
+
+        self.log_dict({'val_loss': loss['total_loss']})
+
+
+    def on_validation_epoch_end(self):
+        if self.metric['total_train_steps']:
+            print('Epoch ', self.current_epoch)
+            epoch_loss = 0
+            epoch_acc = dict(
+                correct_class=0,
+                correct_noobj=0,
+                correct_obj=0,
+                total_class_preds=0,
+                total_noobj=0,
+                total_obj=0
+            )
+            for i in range(self.metric['epoch_train_steps']):
+                lo = self.metric['epoch_train_loss'][i]
+                epoch_loss += lo['total_loss']
+                acc = self.metric['epoch_train_acc'][i]
+                epoch_acc['correct_class'] += acc['correct_class']
+                epoch_acc['correct_noobj'] += acc['correct_noobj']
+                epoch_acc['correct_obj'] += acc['correct_obj']
+                epoch_acc['total_class_preds'] += acc['total_class_preds']
+                epoch_acc['total_noobj'] += acc['total_noobj']
+                epoch_acc['total_obj'] += acc['total_obj']
+
+
+            print("Train -")
+            print(f"Class accuracy is: {(epoch_acc['correct_class']/(epoch_acc['total_class_preds']+1e-16))*100:2f}%")
+            print(f"No obj accuracy is: {(epoch_acc['correct_noobj']/(epoch_acc['total_noobj']+1e-16))*100:2f}%")
+            print(f"Obj accuracy is: {(epoch_acc['correct_obj']/(epoch_acc['total_obj']+1e-16))*100:2f}%")
+            print(f"Total loss: {(epoch_loss/(len(self.metric['epoch_train_loss'])+1e-16)):2f}")
+
+            self.metric['epoch_train_loss'] = []
+            self.metric['epoch_train_acc'] = []
+            self.metric['epoch_train_steps'] = 0
+
+            #---
+            epoch_loss = 0
+            epoch_acc = dict(
+                correct_class=0,
+                correct_noobj=0,
+                correct_obj=0,
+                total_class_preds=0,
+                total_noobj=0,
+                total_obj=0
+            )
+            for i in range(self.metric['epoch_val_steps']):
+                lo = self.metric['epoch_val_loss'][i]
+                epoch_loss += lo['total_loss']
+                acc = self.metric['epoch_val_acc'][i]
+                epoch_acc['correct_class'] += acc['correct_class']
+                epoch_acc['correct_noobj'] += acc['correct_noobj']
+                epoch_acc['correct_obj'] += acc['correct_obj']
+                epoch_acc['total_class_preds'] += acc['total_class_preds']
+                epoch_acc['total_noobj'] += acc['total_noobj']
+                epoch_acc['total_obj'] += acc['total_obj']
+
+            print("Validation -")
+            print(f"Class accuracy is: {(epoch_acc['correct_class']/(epoch_acc['total_class_preds']+1e-16))*100:2f}%")
+            print(f"No obj accuracy is: {(epoch_acc['correct_noobj']/(epoch_acc['total_noobj']+1e-16))*100:2f}%")
+            print(f"Obj accuracy is: {(epoch_acc['correct_obj']/(epoch_acc['total_obj']+1e-16))*100:2f}%")
+            print(f"Total loss: {(epoch_loss/(len(self.metric['epoch_val_loss'])+1e-16)):2f}")
+
+            self.metric['epoch_val_loss'] = []
+            self.metric['epoch_val_acc'] = []
+            self.metric['epoch_val_steps'] = 0
+
+            print("Creating checkpoint...")
+            self.trainer.save_checkpoint(cfg.CHECKPOINT_FILE)
+
+
+    def test_step(self, test_batch, batch_idx):
+        self.validation_step(test_batch, batch_idx)
+
+    def train_dataloader(self):
+        if not self.trainer.train_dataloader:
+            self.trainer.fit_loop.setup_data()
+
+        return self.trainer.train_dataloader
+
+    def configure_optimizers(self):
+        optimizer = torch.optim.Adam(self.parameters(), lr=cfg.LEARNING_RATE, weight_decay=cfg.WEIGHT_DECAY)
+        scheduler = torch.optim.lr_scheduler.OneCycleLR(optimizer,
+                                                  max_lr=cfg.LEARNING_RATE,
+                                                  epochs=self.trainer.max_epochs,
+                                                  steps_per_epoch=len(self.train_dataloader()),
+                                                  pct_start=8 / self.trainer.max_epochs,
+                                                  div_factor=100,
+                                                  final_div_factor=100,
+                                                  three_phase=False,
+                                                  verbose=False
+                                                  )
+        return {
+            "optimizer": optimizer,
+            "lr_scheduler": {
+                "scheduler": scheduler,
+                'interval': 'step', # or 'epoch'
+                'frequency': 1
+            },
+        }
+
+
+    def plot_grad_cam(self, img, target_layers, grad_opacity=1.0):
+        mean, std = [0, 0, 0], [1, 1, 1]
+
+        cam = GradCAM(model=self, target_layers=target_layers)
+
+        transform = transforms.ToTensor()
+        img = transform(img)
+
+        if self.device != img.device:
+            img = img.to(self.device)
+
+        x = img.unsqueeze(0)
+
+        out = self.forward(img)
+        bboxes = []
+        #fig = plt.figure()
+        for i in range(count):
+            plt.subplot(int(count / 5), 5, i + 1)
+            plt.tight_layout()
+
+            targets = [ClassifierOutputTarget(pred_dict['ground_truths'][i].cpu().item())]
+
+            grayscale_cam = cam(input_tensor=pred_dict['images'][i][None, :].cpu(), targets=targets)
+
+            x = denormalize(pred_dict['images'][i].cpu(), mean, std)
+
+            image = np.array(255 * x, np.int16).transpose(1, 2, 0)
+            img_tensor = np.array(x, np.float16).transpose(1, 2, 0)
+
+            visualization = show_cam_on_image(img_tensor, grayscale_cam.transpose(1, 2, 0), use_rgb=True,
+                                              image_weight=(1.0 - grad_opacity) )
+
+            plt.imshow(image, vmin=0, vmax=255)
+            plt.imshow(visualization, vmin=0, vmax=255, alpha=grad_opacity)
+            plt.xticks([])
+            plt.yticks([])
+
+            title = get_data_label_name(pred_dict['ground_truths'][i].item()) + ' / ' + \
+                    get_data_label_name(pred_dict['predicted_vals'][i].item())
+            plt.title(title, fontsize=8)
+
+def sanity_check(model):
+    x = torch.randn((2, 3, cfg.IMAGE_SIZE, cfg.IMAGE_SIZE))
+    out = model(x)
+    assert model(x)[0].shape == (2, 3, cfg.IMAGE_SIZE // 32, cfg.IMAGE_SIZE // 32, cfg.NUM_CLASSES + 5)
+    assert model(x)[1].shape == (2, 3, cfg.IMAGE_SIZE // 16, cfg.IMAGE_SIZE // 16, cfg.NUM_CLASSES + 5)
+    assert model(x)[2].shape == (2, 3, cfg.IMAGE_SIZE // 8, cfg.IMAGE_SIZE // 8, cfg.NUM_CLASSES + 5)
+    print("Success!")

requirements.txt

@@ -0,0 +1,8 @@
+torch
+torch-lr-finder
+torchvision
+pillow
+gradio
+grad-cam
+numpy
+pytorch-lightning

utils.py

@@ -0,0 +1,625 @@
+from typing import List
+
+import cv2
+from torch_lr_finder import LRFinder
+
+import config as cfg
+import matplotlib.pyplot as plt
+import matplotlib.patches as patches
+import numpy as np
+import os
+import random
+import torch
+
+from collections import Counter
+from torch.utils.data import DataLoader
+from tqdm import tqdm
+
+
+
+
+def iou_width_height(boxes1, boxes2):
+    """
+    Parameters:
+        boxes1 (tensor): width and height of the first bounding boxes
+        boxes2 (tensor): width and height of the second bounding boxes
+    Returns:
+        tensor: Intersection over union of the corresponding boxes
+    """
+    intersection = torch.min(boxes1[..., 0], boxes2[..., 0]) * torch.min(
+        boxes1[..., 1], boxes2[..., 1]
+    )
+    union = (
+        boxes1[..., 0] * boxes1[..., 1] + boxes2[..., 0] * boxes2[..., 1] - intersection
+    )
+    return intersection / union
+
+
+def intersection_over_union(boxes_preds, boxes_labels, box_format="midpoint"):
+    """
+    Video explanation of this function:
+    https://youtu.be/XXYG5ZWtjj0
+
+    This function calculates intersection over union (iou) given pred boxes
+    and target boxes.
+
+    Parameters:
+        boxes_preds (tensor): Predictions of Bounding Boxes (BATCH_SIZE, 4)
+        boxes_labels (tensor): Correct labels of Bounding Boxes (BATCH_SIZE, 4)
+        box_format (str): midpoint/corners, if boxes (x,y,w,h) or (x1,y1,x2,y2)
+
+    Returns:
+        tensor: Intersection over union for all examples
+    """
+
+    if box_format == "midpoint":
+        box1_x1 = boxes_preds[..., 0:1] - boxes_preds[..., 2:3] / 2
+        box1_y1 = boxes_preds[..., 1:2] - boxes_preds[..., 3:4] / 2
+        box1_x2 = boxes_preds[..., 0:1] + boxes_preds[..., 2:3] / 2
+        box1_y2 = boxes_preds[..., 1:2] + boxes_preds[..., 3:4] / 2
+        box2_x1 = boxes_labels[..., 0:1] - boxes_labels[..., 2:3] / 2
+        box2_y1 = boxes_labels[..., 1:2] - boxes_labels[..., 3:4] / 2
+        box2_x2 = boxes_labels[..., 0:1] + boxes_labels[..., 2:3] / 2
+        box2_y2 = boxes_labels[..., 1:2] + boxes_labels[..., 3:4] / 2
+
+    if box_format == "corners":
+        box1_x1 = boxes_preds[..., 0:1]
+        box1_y1 = boxes_preds[..., 1:2]
+        box1_x2 = boxes_preds[..., 2:3]
+        box1_y2 = boxes_preds[..., 3:4]
+        box2_x1 = boxes_labels[..., 0:1]
+        box2_y1 = boxes_labels[..., 1:2]
+        box2_x2 = boxes_labels[..., 2:3]
+        box2_y2 = boxes_labels[..., 3:4]
+
+    x1 = torch.max(box1_x1, box2_x1)
+    y1 = torch.max(box1_y1, box2_y1)
+    x2 = torch.min(box1_x2, box2_x2)
+    y2 = torch.min(box1_y2, box2_y2)
+
+    intersection = (x2 - x1).clamp(0) * (y2 - y1).clamp(0)
+    box1_area = abs((box1_x2 - box1_x1) * (box1_y2 - box1_y1))
+    box2_area = abs((box2_x2 - box2_x1) * (box2_y2 - box2_y1))
+
+    return intersection / (box1_area + box2_area - intersection + 1e-6)
+
+
+def non_max_suppression(bboxes, iou_threshold, threshold, box_format="corners"):
+    """
+    Video explanation of this function:
+    https://youtu.be/YDkjWEN8jNA
+
+    Does Non Max Suppression given bboxes
+
+    Parameters:
+        bboxes (list): list of lists containing all bboxes with each bboxes
+        specified as [class_pred, prob_score, x1, y1, x2, y2]
+        iou_threshold (float): threshold where predicted bboxes is correct
+        threshold (float): threshold to remove predicted bboxes (independent of IoU)
+        box_format (str): "midpoint" or "corners" used to specify bboxes
+
+    Returns:
+        list: bboxes after performing NMS given a specific IoU threshold
+    """
+
+    assert type(bboxes) == list
+
+    bboxes = [box for box in bboxes if box[1] > threshold]
+    bboxes = sorted(bboxes, key=lambda x: x[1], reverse=True)
+    bboxes_after_nms = []
+
+    while bboxes:
+        chosen_box = bboxes.pop(0)
+
+        bboxes = [
+            box
+            for box in bboxes
+            if box[0] != chosen_box[0]
+            or intersection_over_union(
+                torch.tensor(chosen_box[2:]),
+                torch.tensor(box[2:]),
+                box_format=box_format,
+            )
+            < iou_threshold
+        ]
+
+        bboxes_after_nms.append(chosen_box)
+
+    return bboxes_after_nms
+
+
+def mean_average_precision(
+    pred_boxes, true_boxes, iou_threshold=0.5, box_format="midpoint", num_classes=20
+):
+    """
+    Video explanation of this function:
+    https://youtu.be/FppOzcDvaDI
+
+    This function calculates mean average precision (mAP)
+
+    Parameters:
+        pred_boxes (list): list of lists containing all bboxes with each bboxes
+        specified as [train_idx, class_prediction, prob_score, x1, y1, x2, y2]
+        true_boxes (list): Similar as pred_boxes except all the correct ones
+        iou_threshold (float): threshold where predicted bboxes is correct
+        box_format (str): "midpoint" or "corners" used to specify bboxes
+        num_classes (int): number of classes
+
+    Returns:
+        float: mAP value across all classes given a specific IoU threshold
+    """
+
+    # list storing all AP for respective classes
+    average_precisions = []
+
+    # used for numerical stability later on
+    epsilon = 1e-6
+
+    for c in range(num_classes):
+        detections = []
+        ground_truths = []
+
+        # Go through all predictions and targets,
+        # and only add the ones that belong to the
+        # current class c
+        for detection in pred_boxes:
+            if detection[1] == c:
+                detections.append(detection)
+
+        for true_box in true_boxes:
+            if true_box[1] == c:
+                ground_truths.append(true_box)
+
+        # find the amount of bboxes for each training example
+        # Counter here finds how many ground truth bboxes we get
+        # for each training example, so let's say img 0 has 3,
+        # img 1 has 5 then we will obtain a dictionary with:
+        # amount_bboxes = {0:3, 1:5}
+        amount_bboxes = Counter([gt[0] for gt in ground_truths])
+
+        # We then go through each key, val in this dictionary
+        # and convert to the following (w.r.t same example):
+        # ammount_bboxes = {0:torch.tensor[0,0,0], 1:torch.tensor[0,0,0,0,0]}
+        for key, val in amount_bboxes.items():
+            amount_bboxes[key] = torch.zeros(val)
+
+        # sort by box probabilities which is index 2
+        detections.sort(key=lambda x: x[2], reverse=True)
+        TP = torch.zeros((len(detections)))
+        FP = torch.zeros((len(detections)))
+        total_true_bboxes = len(ground_truths)
+
+        # If none exists for this class then we can safely skip
+        if total_true_bboxes == 0:
+            continue
+
+        for detection_idx, detection in enumerate(detections):
+            # Only take out the ground_truths that have the same
+            # training idx as detection
+            ground_truth_img = [
+                bbox for bbox in ground_truths if bbox[0] == detection[0]
+            ]
+
+            num_gts = len(ground_truth_img)
+            best_iou = 0
+
+            for idx, gt in enumerate(ground_truth_img):
+                iou = intersection_over_union(
+                    torch.tensor(detection[3:]),
+                    torch.tensor(gt[3:]),
+                    box_format=box_format,
+                )
+
+                if iou > best_iou:
+                    best_iou = iou
+                    best_gt_idx = idx
+
+            if best_iou > iou_threshold:
+                # only detect ground truth detection once
+                if amount_bboxes[detection[0]][best_gt_idx] == 0:
+                    # true positive and add this bounding box to seen
+                    TP[detection_idx] = 1
+                    amount_bboxes[detection[0]][best_gt_idx] = 1
+                else:
+                    FP[detection_idx] = 1
+
+            # if IOU is lower then the detection is a false positive
+            else:
+                FP[detection_idx] = 1
+
+        TP_cumsum = torch.cumsum(TP, dim=0)
+        FP_cumsum = torch.cumsum(FP, dim=0)
+        recalls = TP_cumsum / (total_true_bboxes + epsilon)
+        precisions = TP_cumsum / (TP_cumsum + FP_cumsum + epsilon)
+        precisions = torch.cat((torch.tensor([1]), precisions))
+        recalls = torch.cat((torch.tensor([0]), recalls))
+        # torch.trapz for numerical integration
+        average_precisions.append(torch.trapz(precisions, recalls))
+
+    return sum(average_precisions) / len(average_precisions)
+
+
+def plot_image(image, boxes):
+    """Plots predicted bounding boxes on the image"""
+    cmap = plt.get_cmap("tab20b")
+    class_labels = cfg.COCO_LABELS if cfg.DATASET=='COCO' else cfg.PASCAL_CLASSES
+    colors = [cmap(i) for i in np.linspace(0, 1, len(class_labels))]
+    im = np.array(image)
+    height, width, _ = im.shape
+
+    # Create figure and axes
+    fig, ax = plt.subplots(1)
+    # Display the image
+    ax.imshow(im)
+
+    # box[0] is x midpoint, box[2] is width
+    # box[1] is y midpoint, box[3] is height
+
+    # Create a Rectangle patch
+    for box in boxes:
+        assert len(box) == 6, "box should contain class pred, confidence, x, y, width, height"
+        class_pred = box[0]
+        box = box[2:]
+        upper_left_x = box[0] - box[2] / 2
+        upper_left_y = box[1] - box[3] / 2
+        rect = patches.Rectangle(
+            (upper_left_x * width, upper_left_y * height),
+            box[2] * width,
+            box[3] * height,
+            linewidth=2,
+            edgecolor=colors[int(class_pred)],
+            facecolor="none",
+        )
+        # Add the patch to the Axes
+        ax.add_patch(rect)
+        plt.text(
+            upper_left_x * width,
+            upper_left_y * height,
+            s=class_labels[int(class_pred)],
+            color="white",
+            verticalalignment="top",
+            bbox={"color": colors[int(class_pred)], "pad": 0},
+        )
+
+    plt.show()
+
+
+def get_evaluation_bboxes(
+    loader,
+    model,
+    iou_threshold,
+    anchors,
+    threshold,
+    box_format="midpoint",
+    device="cuda",
+):
+    # make sure model is in eval before get bboxes
+    model.eval()
+    train_idx = 0
+    all_pred_boxes = []
+    all_true_boxes = []
+    for batch_idx, (x, labels) in enumerate(tqdm(loader)):
+        x = x.to(device)
+
+        with torch.no_grad():
+            predictions = model(x)
+
+        batch_size = x.shape[0]
+        bboxes = [[] for _ in range(batch_size)]
+        for i in range(3):
+            S = predictions[i].shape[2]
+            anchor = torch.tensor([*anchors[i]]).to(device) * S
+            boxes_scale_i = cells_to_bboxes(
+                predictions[i], anchor, S=S, is_preds=True
+            )
+            for idx, (box) in enumerate(boxes_scale_i):
+                bboxes[idx] += box
+
+        # we just want one bbox for each label, not one for each scale
+        true_bboxes = cells_to_bboxes(
+            labels[2], anchor, S=S, is_preds=False
+        )
+
+        for idx in range(batch_size):
+            nms_boxes = non_max_suppression(
+                bboxes[idx],
+                iou_threshold=iou_threshold,
+                threshold=threshold,
+                box_format=box_format,
+            )
+
+            for nms_box in nms_boxes:
+                all_pred_boxes.append([train_idx] + nms_box)
+
+            for box in true_bboxes[idx]:
+                if box[1] > threshold:
+                    all_true_boxes.append([train_idx] + box)
+
+            train_idx += 1
+
+    model.train()
+    return all_pred_boxes, all_true_boxes
+
+
+def cells_to_bboxes(predictions, anchors, S, is_preds=True):
+    """
+    Scales the predictions coming from the model to
+    be relative to the entire image such that they for example later
+    can be plotted or.
+    INPUT:
+    predictions: tensor of size (N, 3, S, S, num_classes+5)
+    anchors: the anchors used for the predictions
+    S: the number of cells the image is divided in on the width (and height)
+    is_preds: whether the input is predictions or the true bounding boxes
+    OUTPUT:
+    converted_bboxes: the converted boxes of sizes (N, num_anchors, S, S, 1+5) with class index,
+                      object score, bounding box coordinates
+    """
+    BATCH_SIZE = predictions.shape[0]
+    num_anchors = len(anchors)
+    box_predictions = predictions[..., 1:5]
+    if is_preds:
+        anchors = anchors.reshape(1, len(anchors), 1, 1, 2)
+        box_predictions[..., 0:2] = torch.sigmoid(box_predictions[..., 0:2])
+        box_predictions[..., 2:] = torch.exp(box_predictions[..., 2:]) * anchors
+        scores = torch.sigmoid(predictions[..., 0:1])
+        best_class = torch.argmax(predictions[..., 5:], dim=-1).unsqueeze(-1)
+    else:
+        scores = predictions[..., 0:1]
+        best_class = predictions[..., 5:6]
+
+    cell_indices = (
+        torch.arange(S)
+        .repeat(predictions.shape[0], 3, S, 1)
+        .unsqueeze(-1)
+        .to(predictions.device)
+    )
+    x = 1 / S * (box_predictions[..., 0:1] + cell_indices)
+    y = 1 / S * (box_predictions[..., 1:2] + cell_indices.permute(0, 1, 3, 2, 4))
+    w_h = 1 / S * box_predictions[..., 2:4]
+    converted_bboxes = torch.cat((best_class, scores, x, y, w_h), dim=-1).reshape(BATCH_SIZE, num_anchors * S * S, 6)
+    return converted_bboxes.tolist()
+
+def check_class_accuracy(model, loader, threshold):
+    model.eval()
+    tot_class_preds, correct_class = 0, 0
+    tot_noobj, correct_noobj = 0, 0
+    tot_obj, correct_obj = 0, 0
+
+    for idx, (x, y) in enumerate(tqdm(loader)):
+        x = x.to(cfg.DEVICE)
+        with torch.no_grad():
+            out = model(x)
+
+        for i in range(3):
+            y[i] = y[i].to(cfg.DEVICE)
+            obj = y[i][..., 0] == 1 # in paper this is Iobj_i
+            noobj = y[i][..., 0] == 0  # in paper this is Iobj_i
+
+            correct_class += torch.sum(
+                torch.argmax(out[i][..., 5:][obj], dim=-1) == y[i][..., 5][obj]
+            )
+            tot_class_preds += torch.sum(obj)
+
+            obj_preds = torch.sigmoid(out[i][..., 0]) > threshold
+            correct_obj += torch.sum(obj_preds[obj] == y[i][..., 0][obj])
+            tot_obj += torch.sum(obj)
+            correct_noobj += torch.sum(obj_preds[noobj] == y[i][..., 0][noobj])
+            tot_noobj += torch.sum(noobj)
+
+    print(f"Class accuracy is: {(correct_class/(tot_class_preds+1e-16))*100:2f}%")
+    print(f"No obj accuracy is: {(correct_noobj/(tot_noobj+1e-16))*100:2f}%")
+    print(f"Obj accuracy is: {(correct_obj/(tot_obj+1e-16))*100:2f}%")
+    model.train()
+
+
+def get_mean_std(loader):
+    # var[X] = E[X**2] - E[X]**2
+    channels_sum, channels_sqrd_sum, num_batches = 0, 0, 0
+
+    for data, _ in tqdm(loader):
+        channels_sum += torch.mean(data, dim=[0, 2, 3])
+        channels_sqrd_sum += torch.mean(data ** 2, dim=[0, 2, 3])
+        num_batches += 1
+
+    mean = channels_sum / num_batches
+    std = (channels_sqrd_sum / num_batches - mean ** 2) ** 0.5
+
+    return mean, std
+
+def get_loaders(train_csv_path, test_csv_path):
+    from dataset import YOLODataset
+
+    IMAGE_SIZE = cfg.IMAGE_SIZE
+    train_dataset = YOLODataset(
+        train_csv_path,
+        transform=cfg.train_transforms,
+        S=[IMAGE_SIZE // 32, IMAGE_SIZE // 16, IMAGE_SIZE // 8],
+        img_dir=cfg.IMG_DIR,
+        label_dir=cfg.LABEL_DIR,
+        anchors=cfg.ANCHORS,
+    )
+    test_dataset = YOLODataset(
+        test_csv_path,
+        transform=cfg.test_transforms,
+        S=[IMAGE_SIZE // 32, IMAGE_SIZE // 16, IMAGE_SIZE // 8],
+        img_dir=cfg.IMG_DIR,
+        label_dir=cfg.LABEL_DIR,
+        anchors=cfg.ANCHORS,
+    )
+    train_loader = DataLoader(
+        dataset=train_dataset,
+        batch_size=cfg.BATCH_SIZE,
+        num_workers=cfg.NUM_WORKERS,
+        pin_memory=cfg.PIN_MEMORY,
+        shuffle=True,
+        drop_last=False,
+    )
+    test_loader = DataLoader(
+        dataset=test_dataset,
+        batch_size=cfg.BATCH_SIZE,
+        num_workers=cfg.NUM_WORKERS,
+        pin_memory=cfg.PIN_MEMORY,
+        shuffle=False,
+        drop_last=False,
+    )
+
+    train_eval_dataset = YOLODataset(
+        train_csv_path,
+        transform=cfg.test_transforms,
+        S=[IMAGE_SIZE // 32, IMAGE_SIZE // 16, IMAGE_SIZE // 8],
+        img_dir=cfg.IMG_DIR,
+        label_dir=cfg.LABEL_DIR,
+        anchors=cfg.ANCHORS,
+    )
+    train_eval_loader = DataLoader(
+        dataset=train_eval_dataset,
+        batch_size=cfg.BATCH_SIZE,
+        num_workers=cfg.NUM_WORKERS,
+        pin_memory=cfg.PIN_MEMORY,
+        shuffle=False,
+        drop_last=False,
+    )
+
+    return train_loader, test_loader, train_eval_loader
+
+
+def seed_everything(seed=42):
+    os.environ['PYTHONHASHSEED'] = str(seed)
+    random.seed(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    torch.cuda.manual_seed(seed)
+    torch.cuda.manual_seed_all(seed)
+    torch.backends.cudnn.deterministic = True
+    torch.backends.cudnn.benchmark = False
+
+
+def clip_coords(boxes, img_shape):
+    # Clip bounding xyxy bounding boxes to image shape (height, width)
+    boxes[:, 0].clamp_(0, img_shape[1])  # x1
+    boxes[:, 1].clamp_(0, img_shape[0])  # y1
+    boxes[:, 2].clamp_(0, img_shape[1])  # x2
+    boxes[:, 3].clamp_(0, img_shape[0])  # y2
+
+def xywhn2xyxy(x, w=640, h=640, padw=0, padh=0):
+    # Convert nx4 boxes from [x, y, w, h] normalized to [x1, y1, x2, y2] where xy1=top-left, xy2=bottom-right
+    y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
+    y[..., 0] = w * (x[..., 0] - x[..., 2] / 2) + padw  # top left x
+    y[..., 1] = h * (x[..., 1] - x[..., 3] / 2) + padh  # top left y
+    y[..., 2] = w * (x[..., 0] + x[..., 2] / 2) + padw  # bottom right x
+    y[..., 3] = h * (x[..., 1] + x[..., 3] / 2) + padh  # bottom right y
+    return y
+
+
+def xyn2xy(x, w=640, h=640, padw=0, padh=0):
+    # Convert normalized segments into pixel segments, shape (n,2)
+    y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
+    y[..., 0] = w * x[..., 0] + padw  # top left x
+    y[..., 1] = h * x[..., 1] + padh  # top left y
+    return y
+
+def xyxy2xywhn(x, w=640, h=640, clip=False, eps=0.0):
+    # Convert nx4 boxes from [x1, y1, x2, y2] to [x, y, w, h] normalized where xy1=top-left, xy2=bottom-right
+    if clip:
+        clip_boxes(x, (h - eps, w - eps))  # warning: inplace clip
+    y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
+    y[..., 0] = ((x[..., 0] + x[..., 2]) / 2) / w  # x center
+    y[..., 1] = ((x[..., 1] + x[..., 3]) / 2) / h  # y center
+    y[..., 2] = (x[..., 2] - x[..., 0]) / w  # width
+    y[..., 3] = (x[..., 3] - x[..., 1]) / h  # height
+    return y
+
+def clip_boxes(boxes, shape):
+    # Clip boxes (xyxy) to image shape (height, width)
+    if isinstance(boxes, torch.Tensor):  # faster individually
+        boxes[..., 0].clamp_(0, shape[1])  # x1
+        boxes[..., 1].clamp_(0, shape[0])  # y1
+        boxes[..., 2].clamp_(0, shape[1])  # x2
+        boxes[..., 3].clamp_(0, shape[0])  # y2
+    else:  # np.array (faster grouped)
+        boxes[..., [0, 2]] = boxes[..., [0, 2]].clip(0, shape[1])  # x1, x2
+        boxes[..., [1, 3]] = boxes[..., [1, 3]].clip(0, shape[0])  # y1, y2
+
+
+def find_lr(model, optimizer, criterion, data_loader):
+    lr_finder = LRFinder(model, optimizer, criterion)
+    lr_finder.range_test(data_loader, end_lr=100, num_iter=100)
+    _, best_lr = lr_finder.plot()  # to inspect the loss-learning rate graph
+    lr_finder.reset()
+
+def load_model_from_checkpoint(device, file_name='ckpt_light.pth'):
+    checkpoint = torch.load(file_name, map_location=device)
+
+    return checkpoint
+
+def plot_examples(model, loader, iou_threshold, threshold, anchors):
+
+    print(anchors.device)
+    x, y = next(iter(loader))
+    x = x.to(cfg.DEVICE)
+    out = model(x)
+    bboxes = [[] for _ in range(x.shape[0])]
+    for i in range(3):
+        batch_size, A, S, _, _ = out[i].shape
+        anchor = anchors[i]
+        boxes_scale_i = cells_to_bboxes(
+            out[i], anchor, S=S, is_preds=True
+        )
+        for idx, (box) in enumerate(boxes_scale_i):
+            bboxes[idx] += box
+
+    for i in range(batch_size // 4):
+        nms_boxes = non_max_suppression(
+            bboxes[i], iou_threshold=iou_threshold, threshold=threshold, box_format="midpoint",
+        )
+        plot_image(x[i].permute(1, 2, 0).detach().cpu(), nms_boxes)
+
+
+def draw_predictions(image: np.ndarray, boxes: List[List], class_labels: List[str]) -> np.ndarray:
+    """Plots predicted bounding boxes on the image"""
+
+    colors = [[random.randint(0, 255) for _ in range(3)] for name in class_labels]
+
+    im = np.array(image)
+    height, width, _ = im.shape
+    bbox_thick = 2 #int(0.6 * (height + width) / 600)
+
+    # Create a Rectangle patch
+    for box in boxes:
+        assert len(box) == 6, "box should contain class pred, confidence, x, y, width, height"
+        class_pred = box[0]
+        conf = box[1]
+        box = box[2:]
+        upper_left_x = box[0] - box[2] / 2
+        upper_left_y = box[1] - box[3] / 2
+
+        x1 = int(upper_left_x * width)
+        y1 = int(upper_left_y * height)
+
+        x2 = x1 + int(box[2] * width)
+        y2 = y1 + int(box[3] * height)
+
+        cv2.rectangle(
+            image,
+            (x1, y1), (x2, y2),
+            color=colors[int(class_pred)],
+            thickness=bbox_thick
+        )
+        text = f"{class_labels[int(class_pred)]}" #: {conf:.2f}"
+        t_size = cv2.getTextSize(text, 0, 0.7, thickness=bbox_thick // 2)[0]
+        c3 = (x1 + t_size[0], y1 - t_size[1] - 3)
+
+        cv2.rectangle(image, (x1, y1), c3, colors[int(class_pred)], -1)
+        cv2.putText(
+            image,
+            text,
+            (x1, y1 - 2),
+            cv2.FONT_HERSHEY_SIMPLEX,
+            0.7,
+            (0, 0, 0),
+            bbox_thick // 2,
+            lineType=cv2.LINE_AA,
+        )
+
+    return image