Update

app.py

@@ -1,16 +1,18 @@
-import gradio as gr
-import cv2
+import os
+import torch
+import spaces
 import matplotlib
+
 import numpy as np
-import os
+import gradio as gr
+
 from PIL import Image
-import spaces
-import torch
-import tempfile
-from gradio_imageslider import ImageSlider
+from transformers import pipeline
 from huggingface_hub import hf_hub_download
+from gradio_imageslider import ImageSlider
 
 from depth_anything_v2.dpt import DepthAnythingV2
+from loguru import logger
 
 css = """
 #img-display-container {
@@ -26,78 +28,122 @@ css = """
     height: 62px;
 }
 """
+
+title = "# Depth Anything: Watch V1 and V2 side by side."
+description1 = """Please refer to **Depth Anything V2** [paper](https://arxiv.org/abs/2406.09414) for more details."""
+
 DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'
-model_configs = {
+DEFAULT_V2_MODEL_NAME = "Base"
+DEFAULT_V1_MODEL_NAME = "Base"
+
+cmap = matplotlib.colormaps.get_cmap('Spectral_r')
+
+# --------------------------------------------------------------------
+# Depth anything V1 configuration
+# --------------------------------------------------------------------
+depth_anything_v1_name2checkpoint = {
+    "Small": "LiheYoung/depth-anything-small-hf",
+    "Base": "LiheYoung/depth-anything-base-hf",
+    "Large": "LiheYoung/depth-anything-large-hf",
+}
+
+depth_anything_v1_pipelines = {}
+# --------------------------------------------------------------------
+# Depth anything V2 configuration
+# --------------------------------------------------------------------
+
+depth_anything_v2_configs = {
     'vits': {'encoder': 'vits', 'features': 64, 'out_channels': [48, 96, 192, 384]},
     'vitb': {'encoder': 'vitb', 'features': 128, 'out_channels': [96, 192, 384, 768]},
     'vitl': {'encoder': 'vitl', 'features': 256, 'out_channels': [256, 512, 1024, 1024]},
     'vitg': {'encoder': 'vitg', 'features': 384, 'out_channels': [1536, 1536, 1536, 1536]}
 }
-encoder2name = {
+depth_anything_v2_encoder2name = {
     'vits': 'Small',
     'vitb': 'Base',
     'vitl': 'Large',
-    'vitg': 'Giant', # we are undergoing company review procedures to release our giant model checkpoint
+    # 'vitg': 'Giant', # we are undergoing company review procedures to release our giant model checkpoint
 }
-encoder = 'vitl'
-model_name = encoder2name[encoder]
-model = DepthAnythingV2(**model_configs[encoder])
-filepath = hf_hub_download(repo_id=f"depth-anything/Depth-Anything-V2-{model_name}", filename=f"depth_anything_v2_{encoder}.pth", repo_type="model")
-state_dict = torch.load(filepath, map_location="cpu")
-model.load_state_dict(state_dict)
-model = model.to(DEVICE).eval()
-
-title = "# Depth Anything V2"
-description1 = """Official demo for **Depth Anything V2**.
-Please refer to our [paper](https://arxiv.org/abs/2406.09414) for more details."""
-description2 = """**Due to the issue with our V2 Github repositories, we temporarily upload the content to [Huggingface space](https://huggingface.co/spaces/depth-anything/Depth-Anything-V2/blob/main/README_Github.md).**"""
+depth_anything_v2_name2encoder = {v: k for k, v in depth_anything_v2_encoder2name.items()}
+
+depth_anything_v2_models = {}
+# --------------------------------------------------------------------
+
+
+def get_v1_pipe(model_name):
+    return pipeline(task="depth-estimation", model=depth_anything_v1_name2checkpoint[model_name], device=DEVICE)
+
+
+def get_v2_model(model_name):
+    encoder = depth_anything_v2_name2encoder[model_name]
+    model = DepthAnythingV2(**depth_anything_v2_configs[encoder])
+    filepath = hf_hub_download(repo_id=f"depth-anything/Depth-Anything-V2-{model_name}", filename=f"depth_anything_v2_{encoder}.pth", repo_type="model")
+    state_dict = torch.load(filepath, map_location="cpu")
+    model.load_state_dict(state_dict)
+    model = model.to(DEVICE).eval()
+    return model
+
 
 @spaces.GPU
-def predict_depth(image):
-    return model.infer_image(image)
+def predict_depth_v1(image, model_name):
+    if model_name not in depth_anything_v1_pipelines:
+        depth_anything_v1_pipelines[model_name] = get_v1_pipe(model_name)
+    pipe = depth_anything_v1_pipelines[model_name]
+    return pipe(image)
 
-with gr.Blocks(css=css) as demo:
-    gr.Markdown(title)
-    gr.Markdown(description1)
-    gr.Markdown(description2)
-    gr.Markdown("### Depth Prediction demo")
 
-    with gr.Row():
-        input_image = gr.Image(label="Input Image", type='numpy', elem_id='img-display-input')
-        depth_image_slider = ImageSlider(label="Depth Map with Slider View", elem_id='img-display-output', position=0.5)
-    submit = gr.Button(value="Compute Depth")
-    gray_depth_file = gr.File(label="Grayscale depth map", elem_id="download",)
-    raw_file = gr.File(label="16-bit raw output (can be considered as disparity)", elem_id="download",)
+@spaces.GPU
+def predict_depth_v2(image, model_name):
+    if model_name not in depth_anything_v2_models:
+        depth_anything_v2_models[model_name] = get_v2_model(model_name)
+    model = depth_anything_v2_models[model_name]
+    return model.infer_image(image)
 
-    cmap = matplotlib.colormaps.get_cmap('Spectral_r')
 
-    def on_submit(image):
-        original_image = image.copy()
+def compute_depth_map_v2(image, model_select: str):
+    depth = predict_depth_v2(image[:, :, ::-1], model_select)
+    depth = (depth - depth.min()) / (depth.max() - depth.min()) * 255.0
+    depth = depth.astype(np.uint8)
+    colored_depth = (cmap(depth)[:, :, :3] * 255).astype(np.uint8)
+    return colored_depth
 
-        h, w = image.shape[:2]
 
-        depth = predict_depth(image[:, :, ::-1])
+def compute_depth_map_v1(image, model_select):
+    pil_image = Image.fromarray(image)
+    depth = predict_depth_v1(pil_image, model_select)
+    depth = np.array(depth["depth"]).astype(np.uint8)
+    colored_depth = (cmap(depth)[:, :, :3] * 255).astype(np.uint8)
+    return colored_depth
 
-        raw_depth = Image.fromarray(depth.astype('uint16'))
-        tmp_raw_depth = tempfile.NamedTemporaryFile(suffix='.png', delete=False)
-        raw_depth.save(tmp_raw_depth.name)
 
-        depth = (depth - depth.min()) / (depth.max() - depth.min()) * 255.0
-        depth = depth.astype(np.uint8)
-        colored_depth = (cmap(depth)[:, :, :3] * 255).astype(np.uint8)
+def on_submit(image, model_v1_select, model_v2_select):
+    logger.info(f"Computing depth for V1 model: {model_v1_select} and V2 model: {model_v2_select}")
+    colored_depth_v1 = compute_depth_map_v1(image, model_v1_select)
+    colored_depth_v2 = compute_depth_map_v2(image, model_v2_select)
+    return colored_depth_v1, colored_depth_v2
 
-        gray_depth = Image.fromarray(depth)
-        tmp_gray_depth = tempfile.NamedTemporaryFile(suffix='.png', delete=False)
-        gray_depth.save(tmp_gray_depth.name)
 
-        return [(original_image, colored_depth), tmp_gray_depth.name, tmp_raw_depth.name]
+with gr.Blocks(css=css) as demo:
+    gr.Markdown(title)
+    gr.Markdown(description1)
+    gr.Markdown("### Depth Prediction demo")
+    with gr.Row():
+        model_select_v1 = gr.Dropdown(label="Depth Anything V1 Model", choices=list(depth_anything_v1_name2checkpoint.keys()), value=DEFAULT_V1_MODEL_NAME)
+        model_select_v2 = gr.Dropdown(label="Depth Anything V2 Model", choices=list(depth_anything_v2_encoder2name.values()), value=DEFAULT_V2_MODEL_NAME)
+    with gr.Row():
+        gr.Markdown()
+        gr.Markdown("Depth Maps: V1 <-> V2")
+    with gr.Row():
+        input_image = gr.Image(label="Input Image", type='numpy', elem_id='img-display-input')
+        depth_image_slider = ImageSlider(elem_id='img-display-output', position=0.5)
 
-    submit.click(on_submit, inputs=[input_image], outputs=[depth_image_slider, gray_depth_file, raw_file])
+    submit = gr.Button(value="Compute Depth")
+    submit.click(on_submit, inputs=[input_image, model_select_v1, model_select_v2], outputs=[depth_image_slider])
 
     example_files = os.listdir('assets/examples')
     example_files.sort()
     example_files = [os.path.join('assets/examples', filename) for filename in example_files]
-    examples = gr.Examples(examples=example_files, inputs=[input_image], outputs=[depth_image_slider, gray_depth_file, raw_file], fn=on_submit)
+    examples = gr.Examples(examples=example_files, inputs=[input_image], outputs=[depth_image_slider], fn=on_submit)
 
 
 if __name__ == '__main__':

requirements.txt

@@ -4,4 +4,7 @@ torch
 torchvision
 opencv-python
 matplotlib
-huggingface_hub
+huggingface_hub
+transformers
+numpy==1.*
+loguru