add post controlnet

.gitattributes

@@ -39,3 +39,5 @@ image/woman_1.png filter=lfs diff=lfs merge=lfs -text
 image/woman_2.png filter=lfs diff=lfs merge=lfs -text
 image/1.png filter=lfs diff=lfs merge=lfs -text
 image/2.png filter=lfs diff=lfs merge=lfs -text
+image/woman_4.png filter=lfs diff=lfs merge=lfs -text
+image/woman_3.png filter=lfs diff=lfs merge=lfs -text

annotator/ckpts/dw-ll_ucoco_384.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:724f4ff2439ed61afb86fb8a1951ec39c6220682803b4a8bd4f598cd913b1843
+size 134399116

annotator/ckpts/yolox_l.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7860ae79de6c89a3c1eb72ae9a2756c0ccfbe04b7791bb5880afabd97855a411
+size 216746733

annotator/dwpose/__init__.py

@@ -0,0 +1,91 @@
+# Openpose
+# Original from CMU https://github.com/CMU-Perceptual-Computing-Lab/openpose
+# 2nd Edited by https://github.com/Hzzone/pytorch-openpose
+# 3rd Edited by ControlNet
+# 4th Edited by ControlNet (added face and correct hands)
+
+import os
+os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
+
+import torch
+import numpy as np
+from . import util
+from .wholebody import Wholebody
+
+def draw_pose(pose, H, W):
+    bodies = pose['bodies']
+    faces = pose['faces']
+    hands = pose['hands']
+    candidate = bodies['candidate']
+    subset = bodies['subset']
+    canvas = np.zeros(shape=(H, W, 3), dtype=np.uint8)
+
+    canvas = util.draw_bodypose(canvas, candidate, subset)
+
+    canvas = util.draw_handpose(canvas, hands)
+
+    # canvas = util.draw_facepose(canvas, faces)
+
+    return canvas
+
+
+class DWposeDetector:
+    def __init__(self):
+
+        self.pose_estimation = Wholebody()
+
+    
+    def getres(self, oriImg):
+        out_res = {}
+        oriImg = oriImg.copy()
+        H, W, C = oriImg.shape
+        with torch.no_grad():
+            candidate, subset = self.pose_estimation(oriImg)
+            out_res['candidate']=candidate
+            out_res['subset']=subset
+            out_res['width']=W
+            out_res['height']=H
+            return out_res
+
+    def __call__(self, oriImg):
+
+        oriImg = oriImg.copy()
+        H, W, C = oriImg.shape
+        with torch.no_grad():
+            _candidate, _subset = self.pose_estimation(oriImg)
+            
+            subset = _subset.copy()
+            candidate = _candidate.copy()
+            nums, keys, locs = candidate.shape
+            candidate[..., 0] /= float(W)
+            candidate[..., 1] /= float(H)
+            body = candidate[:,:18].copy()
+            body = body.reshape(nums*18, locs)
+            score = subset[:,:18]
+            for i in range(len(score)):
+                for j in range(len(score[i])):
+                    if score[i][j] > 0.3:
+                        score[i][j] = int(18*i+j)
+                    else:
+                        score[i][j] = -1
+
+            un_visible = subset<0.3
+            candidate[un_visible] = -1
+
+            foot = candidate[:,18:24]
+
+            faces = candidate[:,24:92]
+
+            hands = candidate[:,92:113]
+            hands = np.vstack([hands, candidate[:,113:]])
+            
+            bodies = dict(candidate=body, subset=score)
+            pose = dict(bodies=bodies, hands=hands, faces=faces)
+
+            out_res = {}
+            out_res['candidate']=candidate
+            out_res['subset']=subset
+            out_res['width']=W
+            out_res['height']=H
+
+            return out_res,draw_pose(pose, H, W)

annotator/dwpose/onnxdet.py

@@ -0,0 +1,125 @@
+import cv2
+import numpy as np
+
+import onnxruntime
+
+def nms(boxes, scores, nms_thr):
+    """Single class NMS implemented in Numpy."""
+    x1 = boxes[:, 0]
+    y1 = boxes[:, 1]
+    x2 = boxes[:, 2]
+    y2 = boxes[:, 3]
+
+    areas = (x2 - x1 + 1) * (y2 - y1 + 1)
+    order = scores.argsort()[::-1]
+
+    keep = []
+    while order.size > 0:
+        i = order[0]
+        keep.append(i)
+        xx1 = np.maximum(x1[i], x1[order[1:]])
+        yy1 = np.maximum(y1[i], y1[order[1:]])
+        xx2 = np.minimum(x2[i], x2[order[1:]])
+        yy2 = np.minimum(y2[i], y2[order[1:]])
+
+        w = np.maximum(0.0, xx2 - xx1 + 1)
+        h = np.maximum(0.0, yy2 - yy1 + 1)
+        inter = w * h
+        ovr = inter / (areas[i] + areas[order[1:]] - inter)
+
+        inds = np.where(ovr <= nms_thr)[0]
+        order = order[inds + 1]
+
+    return keep
+
+def multiclass_nms(boxes, scores, nms_thr, score_thr):
+    """Multiclass NMS implemented in Numpy. Class-aware version."""
+    final_dets = []
+    num_classes = scores.shape[1]
+    for cls_ind in range(num_classes):
+        cls_scores = scores[:, cls_ind]
+        valid_score_mask = cls_scores > score_thr
+        if valid_score_mask.sum() == 0:
+            continue
+        else:
+            valid_scores = cls_scores[valid_score_mask]
+            valid_boxes = boxes[valid_score_mask]
+            keep = nms(valid_boxes, valid_scores, nms_thr)
+            if len(keep) > 0:
+                cls_inds = np.ones((len(keep), 1)) * cls_ind
+                dets = np.concatenate(
+                    [valid_boxes[keep], valid_scores[keep, None], cls_inds], 1
+                )
+                final_dets.append(dets)
+    if len(final_dets) == 0:
+        return None
+    return np.concatenate(final_dets, 0)
+
+def demo_postprocess(outputs, img_size, p6=False):
+    grids = []
+    expanded_strides = []
+    strides = [8, 16, 32] if not p6 else [8, 16, 32, 64]
+
+    hsizes = [img_size[0] // stride for stride in strides]
+    wsizes = [img_size[1] // stride for stride in strides]
+
+    for hsize, wsize, stride in zip(hsizes, wsizes, strides):
+        xv, yv = np.meshgrid(np.arange(wsize), np.arange(hsize))
+        grid = np.stack((xv, yv), 2).reshape(1, -1, 2)
+        grids.append(grid)
+        shape = grid.shape[:2]
+        expanded_strides.append(np.full((*shape, 1), stride))
+
+    grids = np.concatenate(grids, 1)
+    expanded_strides = np.concatenate(expanded_strides, 1)
+    outputs[..., :2] = (outputs[..., :2] + grids) * expanded_strides
+    outputs[..., 2:4] = np.exp(outputs[..., 2:4]) * expanded_strides
+
+    return outputs
+
+def preprocess(img, input_size, swap=(2, 0, 1)):
+    if len(img.shape) == 3:
+        padded_img = np.ones((input_size[0], input_size[1], 3), dtype=np.uint8) * 114
+    else:
+        padded_img = np.ones(input_size, dtype=np.uint8) * 114
+
+    r = min(input_size[0] / img.shape[0], input_size[1] / img.shape[1])
+    resized_img = cv2.resize(
+        img,
+        (int(img.shape[1] * r), int(img.shape[0] * r)),
+        interpolation=cv2.INTER_LINEAR,
+    ).astype(np.uint8)
+    padded_img[: int(img.shape[0] * r), : int(img.shape[1] * r)] = resized_img
+
+    padded_img = padded_img.transpose(swap)
+    padded_img = np.ascontiguousarray(padded_img, dtype=np.float32)
+    return padded_img, r
+
+def inference_detector(session, oriImg):
+    input_shape = (640,640)
+    img, ratio = preprocess(oriImg, input_shape)
+
+    ort_inputs = {session.get_inputs()[0].name: img[None, :, :, :]}
+    output = session.run(None, ort_inputs)
+    predictions = demo_postprocess(output[0], input_shape)[0]
+
+    boxes = predictions[:, :4]
+    scores = predictions[:, 4:5] * predictions[:, 5:]
+
+    boxes_xyxy = np.ones_like(boxes)
+    boxes_xyxy[:, 0] = boxes[:, 0] - boxes[:, 2]/2.
+    boxes_xyxy[:, 1] = boxes[:, 1] - boxes[:, 3]/2.
+    boxes_xyxy[:, 2] = boxes[:, 0] + boxes[:, 2]/2.
+    boxes_xyxy[:, 3] = boxes[:, 1] + boxes[:, 3]/2.
+    boxes_xyxy /= ratio
+    dets = multiclass_nms(boxes_xyxy, scores, nms_thr=0.45, score_thr=0.1)
+    if dets is not None:
+        final_boxes, final_scores, final_cls_inds = dets[:, :4], dets[:, 4], dets[:, 5]
+        isscore = final_scores>0.3
+        iscat = final_cls_inds == 0
+        isbbox = [ i and j for (i, j) in zip(isscore, iscat)]
+        final_boxes = final_boxes[isbbox]
+    else:
+        final_boxes = np.array([])
+
+    return final_boxes

annotator/dwpose/onnxpose.py

@@ -0,0 +1,360 @@
+from typing import List, Tuple
+
+import cv2
+import numpy as np
+import onnxruntime as ort
+
+def preprocess(
+    img: np.ndarray, out_bbox, input_size: Tuple[int, int] = (192, 256)
+) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
+    """Do preprocessing for RTMPose model inference.
+
+    Args:
+        img (np.ndarray): Input image in shape.
+        input_size (tuple): Input image size in shape (w, h).
+
+    Returns:
+        tuple:
+        - resized_img (np.ndarray): Preprocessed image.
+        - center (np.ndarray): Center of image.
+        - scale (np.ndarray): Scale of image.
+    """
+    # get shape of image
+    img_shape = img.shape[:2]
+    out_img, out_center, out_scale = [], [], []
+    if len(out_bbox) == 0:
+        out_bbox = [[0, 0, img_shape[1], img_shape[0]]]
+    for i in range(len(out_bbox)):
+        x0 = out_bbox[i][0]
+        y0 = out_bbox[i][1]
+        x1 = out_bbox[i][2]
+        y1 = out_bbox[i][3]
+        bbox = np.array([x0, y0, x1, y1])
+
+        # get center and scale
+        center, scale = bbox_xyxy2cs(bbox, padding=1.25)
+
+        # do affine transformation
+        resized_img, scale = top_down_affine(input_size, scale, center, img)
+
+        # normalize image
+        mean = np.array([123.675, 116.28, 103.53])
+        std = np.array([58.395, 57.12, 57.375])
+        resized_img = (resized_img - mean) / std
+
+        out_img.append(resized_img)
+        out_center.append(center)
+        out_scale.append(scale)
+
+    return out_img, out_center, out_scale
+
+
+def inference(sess: ort.InferenceSession, img: np.ndarray) -> np.ndarray:
+    """Inference RTMPose model.
+
+    Args:
+        sess (ort.InferenceSession): ONNXRuntime session.
+        img (np.ndarray): Input image in shape.
+
+    Returns:
+        outputs (np.ndarray): Output of RTMPose model.
+    """
+    all_out = []
+    # build input
+    for i in range(len(img)):
+        input = [img[i].transpose(2, 0, 1)]
+
+        # build output
+        sess_input = {sess.get_inputs()[0].name: input}
+        sess_output = []
+        for out in sess.get_outputs():
+            sess_output.append(out.name)
+
+        # run model
+        outputs = sess.run(sess_output, sess_input)
+        all_out.append(outputs)
+
+    return all_out
+
+
+def postprocess(outputs: List[np.ndarray],
+                model_input_size: Tuple[int, int],
+                center: Tuple[int, int],
+                scale: Tuple[int, int],
+                simcc_split_ratio: float = 2.0
+                ) -> Tuple[np.ndarray, np.ndarray]:
+    """Postprocess for RTMPose model output.
+
+    Args:
+        outputs (np.ndarray): Output of RTMPose model.
+        model_input_size (tuple): RTMPose model Input image size.
+        center (tuple): Center of bbox in shape (x, y).
+        scale (tuple): Scale of bbox in shape (w, h).
+        simcc_split_ratio (float): Split ratio of simcc.
+
+    Returns:
+        tuple:
+        - keypoints (np.ndarray): Rescaled keypoints.
+        - scores (np.ndarray): Model predict scores.
+    """
+    all_key = []
+    all_score = []
+    for i in range(len(outputs)):
+        # use simcc to decode
+        simcc_x, simcc_y = outputs[i]
+        keypoints, scores = decode(simcc_x, simcc_y, simcc_split_ratio)
+
+        # rescale keypoints
+        keypoints = keypoints / model_input_size * scale[i] + center[i] - scale[i] / 2
+        all_key.append(keypoints[0])
+        all_score.append(scores[0])
+
+    return np.array(all_key), np.array(all_score)
+
+
+def bbox_xyxy2cs(bbox: np.ndarray,
+                 padding: float = 1.) -> Tuple[np.ndarray, np.ndarray]:
+    """Transform the bbox format from (x,y,w,h) into (center, scale)
+
+    Args:
+        bbox (ndarray): Bounding box(es) in shape (4,) or (n, 4), formatted
+            as (left, top, right, bottom)
+        padding (float): BBox padding factor that will be multilied to scale.
+            Default: 1.0
+
+    Returns:
+        tuple: A tuple containing center and scale.
+        - np.ndarray[float32]: Center (x, y) of the bbox in shape (2,) or
+            (n, 2)
+        - np.ndarray[float32]: Scale (w, h) of the bbox in shape (2,) or
+            (n, 2)
+    """
+    # convert single bbox from (4, ) to (1, 4)
+    dim = bbox.ndim
+    if dim == 1:
+        bbox = bbox[None, :]
+
+    # get bbox center and scale
+    x1, y1, x2, y2 = np.hsplit(bbox, [1, 2, 3])
+    center = np.hstack([x1 + x2, y1 + y2]) * 0.5
+    scale = np.hstack([x2 - x1, y2 - y1]) * padding
+
+    if dim == 1:
+        center = center[0]
+        scale = scale[0]
+
+    return center, scale
+
+
+def _fix_aspect_ratio(bbox_scale: np.ndarray,
+                      aspect_ratio: float) -> np.ndarray:
+    """Extend the scale to match the given aspect ratio.
+
+    Args:
+        scale (np.ndarray): The image scale (w, h) in shape (2, )
+        aspect_ratio (float): The ratio of ``w/h``
+
+    Returns:
+        np.ndarray: The reshaped image scale in (2, )
+    """
+    w, h = np.hsplit(bbox_scale, [1])
+    bbox_scale = np.where(w > h * aspect_ratio,
+                          np.hstack([w, w / aspect_ratio]),
+                          np.hstack([h * aspect_ratio, h]))
+    return bbox_scale
+
+
+def _rotate_point(pt: np.ndarray, angle_rad: float) -> np.ndarray:
+    """Rotate a point by an angle.
+
+    Args:
+        pt (np.ndarray): 2D point coordinates (x, y) in shape (2, )
+        angle_rad (float): rotation angle in radian
+
+    Returns:
+        np.ndarray: Rotated point in shape (2, )
+    """
+    sn, cs = np.sin(angle_rad), np.cos(angle_rad)
+    rot_mat = np.array([[cs, -sn], [sn, cs]])
+    return rot_mat @ pt
+
+
+def _get_3rd_point(a: np.ndarray, b: np.ndarray) -> np.ndarray:
+    """To calculate the affine matrix, three pairs of points are required. This
+    function is used to get the 3rd point, given 2D points a & b.
+
+    The 3rd point is defined by rotating vector `a - b` by 90 degrees
+    anticlockwise, using b as the rotation center.
+
+    Args:
+        a (np.ndarray): The 1st point (x,y) in shape (2, )
+        b (np.ndarray): The 2nd point (x,y) in shape (2, )
+
+    Returns:
+        np.ndarray: The 3rd point.
+    """
+    direction = a - b
+    c = b + np.r_[-direction[1], direction[0]]
+    return c
+
+
+def get_warp_matrix(center: np.ndarray,
+                    scale: np.ndarray,
+                    rot: float,
+                    output_size: Tuple[int, int],
+                    shift: Tuple[float, float] = (0., 0.),
+                    inv: bool = False) -> np.ndarray:
+    """Calculate the affine transformation matrix that can warp the bbox area
+    in the input image to the output size.
+
+    Args:
+        center (np.ndarray[2, ]): Center of the bounding box (x, y).
+        scale (np.ndarray[2, ]): Scale of the bounding box
+            wrt [width, height].
+        rot (float): Rotation angle (degree).
+        output_size (np.ndarray[2, ] | list(2,)): Size of the
+            destination heatmaps.
+        shift (0-100%): Shift translation ratio wrt the width/height.
+            Default (0., 0.).
+        inv (bool): Option to inverse the affine transform direction.
+            (inv=False: src->dst or inv=True: dst->src)
+
+    Returns:
+        np.ndarray: A 2x3 transformation matrix
+    """
+    shift = np.array(shift)
+    src_w = scale[0]
+    dst_w = output_size[0]
+    dst_h = output_size[1]
+
+    # compute transformation matrix
+    rot_rad = np.deg2rad(rot)
+    src_dir = _rotate_point(np.array([0., src_w * -0.5]), rot_rad)
+    dst_dir = np.array([0., dst_w * -0.5])
+
+    # get four corners of the src rectangle in the original image
+    src = np.zeros((3, 2), dtype=np.float32)
+    src[0, :] = center + scale * shift
+    src[1, :] = center + src_dir + scale * shift
+    src[2, :] = _get_3rd_point(src[0, :], src[1, :])
+
+    # get four corners of the dst rectangle in the input image
+    dst = np.zeros((3, 2), dtype=np.float32)
+    dst[0, :] = [dst_w * 0.5, dst_h * 0.5]
+    dst[1, :] = np.array([dst_w * 0.5, dst_h * 0.5]) + dst_dir
+    dst[2, :] = _get_3rd_point(dst[0, :], dst[1, :])
+
+    if inv:
+        warp_mat = cv2.getAffineTransform(np.float32(dst), np.float32(src))
+    else:
+        warp_mat = cv2.getAffineTransform(np.float32(src), np.float32(dst))
+
+    return warp_mat
+
+
+def top_down_affine(input_size: dict, bbox_scale: dict, bbox_center: dict,
+                    img: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
+    """Get the bbox image as the model input by affine transform.
+
+    Args:
+        input_size (dict): The input size of the model.
+        bbox_scale (dict): The bbox scale of the img.
+        bbox_center (dict): The bbox center of the img.
+        img (np.ndarray): The original image.
+
+    Returns:
+        tuple: A tuple containing center and scale.
+        - np.ndarray[float32]: img after affine transform.
+        - np.ndarray[float32]: bbox scale after affine transform.
+    """
+    w, h = input_size
+    warp_size = (int(w), int(h))
+
+    # reshape bbox to fixed aspect ratio
+    bbox_scale = _fix_aspect_ratio(bbox_scale, aspect_ratio=w / h)
+
+    # get the affine matrix
+    center = bbox_center
+    scale = bbox_scale
+    rot = 0
+    warp_mat = get_warp_matrix(center, scale, rot, output_size=(w, h))
+
+    # do affine transform
+    img = cv2.warpAffine(img, warp_mat, warp_size, flags=cv2.INTER_LINEAR)
+
+    return img, bbox_scale
+
+
+def get_simcc_maximum(simcc_x: np.ndarray,
+                      simcc_y: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
+    """Get maximum response location and value from simcc representations.
+
+    Note:
+        instance number: N
+        num_keypoints: K
+        heatmap height: H
+        heatmap width: W
+
+    Args:
+        simcc_x (np.ndarray): x-axis SimCC in shape (K, Wx) or (N, K, Wx)
+        simcc_y (np.ndarray): y-axis SimCC in shape (K, Wy) or (N, K, Wy)
+
+    Returns:
+        tuple:
+        - locs (np.ndarray): locations of maximum heatmap responses in shape
+            (K, 2) or (N, K, 2)
+        - vals (np.ndarray): values of maximum heatmap responses in shape
+            (K,) or (N, K)
+    """
+    N, K, Wx = simcc_x.shape
+    simcc_x = simcc_x.reshape(N * K, -1)
+    simcc_y = simcc_y.reshape(N * K, -1)
+
+    # get maximum value locations
+    x_locs = np.argmax(simcc_x, axis=1)
+    y_locs = np.argmax(simcc_y, axis=1)
+    locs = np.stack((x_locs, y_locs), axis=-1).astype(np.float32)
+    max_val_x = np.amax(simcc_x, axis=1)
+    max_val_y = np.amax(simcc_y, axis=1)
+
+    # get maximum value across x and y axis
+    mask = max_val_x > max_val_y
+    max_val_x[mask] = max_val_y[mask]
+    vals = max_val_x
+    locs[vals <= 0.] = -1
+
+    # reshape
+    locs = locs.reshape(N, K, 2)
+    vals = vals.reshape(N, K)
+
+    return locs, vals
+
+
+def decode(simcc_x: np.ndarray, simcc_y: np.ndarray,
+           simcc_split_ratio) -> Tuple[np.ndarray, np.ndarray]:
+    """Modulate simcc distribution with Gaussian.
+
+    Args:
+        simcc_x (np.ndarray[K, Wx]): model predicted simcc in x.
+        simcc_y (np.ndarray[K, Wy]): model predicted simcc in y.
+        simcc_split_ratio (int): The split ratio of simcc.
+
+    Returns:
+        tuple: A tuple containing center and scale.
+        - np.ndarray[float32]: keypoints in shape (K, 2) or (n, K, 2)
+        - np.ndarray[float32]: scores in shape (K,) or (n, K)
+    """
+    keypoints, scores = get_simcc_maximum(simcc_x, simcc_y)
+    keypoints /= simcc_split_ratio
+
+    return keypoints, scores
+
+
+def inference_pose(session, out_bbox, oriImg):
+    h, w = session.get_inputs()[0].shape[2:]
+    model_input_size = (w, h)
+    resized_img, center, scale = preprocess(oriImg, out_bbox, model_input_size)
+    outputs = inference(session, resized_img)
+    keypoints, scores = postprocess(outputs, model_input_size, center, scale)
+
+    return keypoints, scores

annotator/dwpose/util.py

@@ -0,0 +1,297 @@
+import math
+import numpy as np
+import matplotlib
+import cv2
+
+
+eps = 0.01
+
+
+def smart_resize(x, s):
+    Ht, Wt = s
+    if x.ndim == 2:
+        Ho, Wo = x.shape
+        Co = 1
+    else:
+        Ho, Wo, Co = x.shape
+    if Co == 3 or Co == 1:
+        k = float(Ht + Wt) / float(Ho + Wo)
+        return cv2.resize(x, (int(Wt), int(Ht)), interpolation=cv2.INTER_AREA if k < 1 else cv2.INTER_LANCZOS4)
+    else:
+        return np.stack([smart_resize(x[:, :, i], s) for i in range(Co)], axis=2)
+
+
+def smart_resize_k(x, fx, fy):
+    if x.ndim == 2:
+        Ho, Wo = x.shape
+        Co = 1
+    else:
+        Ho, Wo, Co = x.shape
+    Ht, Wt = Ho * fy, Wo * fx
+    if Co == 3 or Co == 1:
+        k = float(Ht + Wt) / float(Ho + Wo)
+        return cv2.resize(x, (int(Wt), int(Ht)), interpolation=cv2.INTER_AREA if k < 1 else cv2.INTER_LANCZOS4)
+    else:
+        return np.stack([smart_resize_k(x[:, :, i], fx, fy) for i in range(Co)], axis=2)
+
+
+def padRightDownCorner(img, stride, padValue):
+    h = img.shape[0]
+    w = img.shape[1]
+
+    pad = 4 * [None]
+    pad[0] = 0 # up
+    pad[1] = 0 # left
+    pad[2] = 0 if (h % stride == 0) else stride - (h % stride) # down
+    pad[3] = 0 if (w % stride == 0) else stride - (w % stride) # right
+
+    img_padded = img
+    pad_up = np.tile(img_padded[0:1, :, :]*0 + padValue, (pad[0], 1, 1))
+    img_padded = np.concatenate((pad_up, img_padded), axis=0)
+    pad_left = np.tile(img_padded[:, 0:1, :]*0 + padValue, (1, pad[1], 1))
+    img_padded = np.concatenate((pad_left, img_padded), axis=1)
+    pad_down = np.tile(img_padded[-2:-1, :, :]*0 + padValue, (pad[2], 1, 1))
+    img_padded = np.concatenate((img_padded, pad_down), axis=0)
+    pad_right = np.tile(img_padded[:, -2:-1, :]*0 + padValue, (1, pad[3], 1))
+    img_padded = np.concatenate((img_padded, pad_right), axis=1)
+
+    return img_padded, pad
+
+
+def transfer(model, model_weights):
+    transfered_model_weights = {}
+    for weights_name in model.state_dict().keys():
+        transfered_model_weights[weights_name] = model_weights['.'.join(weights_name.split('.')[1:])]
+    return transfered_model_weights
+
+
+def draw_bodypose(canvas, candidate, subset):
+    H, W, C = canvas.shape
+    candidate = np.array(candidate)
+    subset = np.array(subset)
+
+    stickwidth = 4
+
+    limbSeq = [[2, 3], [2, 6], [3, 4], [4, 5], [6, 7], [7, 8], [2, 9], [9, 10], \
+               [10, 11], [2, 12], [12, 13], [13, 14], [2, 1], [1, 15], [15, 17], \
+               [1, 16], [16, 18], [3, 17], [6, 18]]
+
+    colors = [[255, 0, 0], [255, 85, 0], [255, 170, 0], [255, 255, 0], [170, 255, 0], [85, 255, 0], [0, 255, 0], \
+              [0, 255, 85], [0, 255, 170], [0, 255, 255], [0, 170, 255], [0, 85, 255], [0, 0, 255], [85, 0, 255], \
+              [170, 0, 255], [255, 0, 255], [255, 0, 170], [255, 0, 85]]
+
+    for i in range(17):
+        for n in range(len(subset)):
+            index = subset[n][np.array(limbSeq[i]) - 1]
+            if -1 in index:
+                continue
+            Y = candidate[index.astype(int), 0] * float(W)
+            X = candidate[index.astype(int), 1] * float(H)
+            mX = np.mean(X)
+            mY = np.mean(Y)
+            length = ((X[0] - X[1]) ** 2 + (Y[0] - Y[1]) ** 2) ** 0.5
+            angle = math.degrees(math.atan2(X[0] - X[1], Y[0] - Y[1]))
+            polygon = cv2.ellipse2Poly((int(mY), int(mX)), (int(length / 2), stickwidth), int(angle), 0, 360, 1)
+            cv2.fillConvexPoly(canvas, polygon, colors[i])
+
+    canvas = (canvas * 0.6).astype(np.uint8)
+
+    for i in range(18):
+        for n in range(len(subset)):
+            index = int(subset[n][i])
+            if index == -1:
+                continue
+            x, y = candidate[index][0:2]
+            x = int(x * W)
+            y = int(y * H)
+            cv2.circle(canvas, (int(x), int(y)), 4, colors[i], thickness=-1)
+
+    return canvas
+
+
+def draw_handpose(canvas, all_hand_peaks):
+    H, W, C = canvas.shape
+
+    edges = [[0, 1], [1, 2], [2, 3], [3, 4], [0, 5], [5, 6], [6, 7], [7, 8], [0, 9], [9, 10], \
+             [10, 11], [11, 12], [0, 13], [13, 14], [14, 15], [15, 16], [0, 17], [17, 18], [18, 19], [19, 20]]
+
+    for peaks in all_hand_peaks:
+        peaks = np.array(peaks)
+
+        for ie, e in enumerate(edges):
+            x1, y1 = peaks[e[0]]
+            x2, y2 = peaks[e[1]]
+            x1 = int(x1 * W)
+            y1 = int(y1 * H)
+            x2 = int(x2 * W)
+            y2 = int(y2 * H)
+            if x1 > eps and y1 > eps and x2 > eps and y2 > eps:
+                cv2.line(canvas, (x1, y1), (x2, y2), matplotlib.colors.hsv_to_rgb([ie / float(len(edges)), 1.0, 1.0]) * 255, thickness=2)
+
+        for i, keyponit in enumerate(peaks):
+            x, y = keyponit
+            x = int(x * W)
+            y = int(y * H)
+            if x > eps and y > eps:
+                cv2.circle(canvas, (x, y), 4, (0, 0, 255), thickness=-1)
+    return canvas
+
+
+def draw_facepose(canvas, all_lmks):
+    H, W, C = canvas.shape
+    for lmks in all_lmks:
+        lmks = np.array(lmks)
+        for lmk in lmks:
+            x, y = lmk
+            x = int(x * W)
+            y = int(y * H)
+            if x > eps and y > eps:
+                cv2.circle(canvas, (x, y), 3, (255, 255, 255), thickness=-1)
+    return canvas
+
+
+# detect hand according to body pose keypoints
+# please refer to https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/src/openpose/hand/handDetector.cpp
+def handDetect(candidate, subset, oriImg):
+    # right hand: wrist 4, elbow 3, shoulder 2
+    # left hand: wrist 7, elbow 6, shoulder 5
+    ratioWristElbow = 0.33
+    detect_result = []
+    image_height, image_width = oriImg.shape[0:2]
+    for person in subset.astype(int):
+        # if any of three not detected
+        has_left = np.sum(person[[5, 6, 7]] == -1) == 0
+        has_right = np.sum(person[[2, 3, 4]] == -1) == 0
+        if not (has_left or has_right):
+            continue
+        hands = []
+        #left hand
+        if has_left:
+            left_shoulder_index, left_elbow_index, left_wrist_index = person[[5, 6, 7]]
+            x1, y1 = candidate[left_shoulder_index][:2]
+            x2, y2 = candidate[left_elbow_index][:2]
+            x3, y3 = candidate[left_wrist_index][:2]
+            hands.append([x1, y1, x2, y2, x3, y3, True])
+        # right hand
+        if has_right:
+            right_shoulder_index, right_elbow_index, right_wrist_index = person[[2, 3, 4]]
+            x1, y1 = candidate[right_shoulder_index][:2]
+            x2, y2 = candidate[right_elbow_index][:2]
+            x3, y3 = candidate[right_wrist_index][:2]
+            hands.append([x1, y1, x2, y2, x3, y3, False])
+
+        for x1, y1, x2, y2, x3, y3, is_left in hands:
+            # pos_hand = pos_wrist + ratio * (pos_wrist - pos_elbox) = (1 + ratio) * pos_wrist - ratio * pos_elbox
+            # handRectangle.x = posePtr[wrist*3] + ratioWristElbow * (posePtr[wrist*3] - posePtr[elbow*3]);
+            # handRectangle.y = posePtr[wrist*3+1] + ratioWristElbow * (posePtr[wrist*3+1] - posePtr[elbow*3+1]);
+            # const auto distanceWristElbow = getDistance(poseKeypoints, person, wrist, elbow);
+            # const auto distanceElbowShoulder = getDistance(poseKeypoints, person, elbow, shoulder);
+            # handRectangle.width = 1.5f * fastMax(distanceWristElbow, 0.9f * distanceElbowShoulder);
+            x = x3 + ratioWristElbow * (x3 - x2)
+            y = y3 + ratioWristElbow * (y3 - y2)
+            distanceWristElbow = math.sqrt((x3 - x2) ** 2 + (y3 - y2) ** 2)
+            distanceElbowShoulder = math.sqrt((x2 - x1) ** 2 + (y2 - y1) ** 2)
+            width = 1.5 * max(distanceWristElbow, 0.9 * distanceElbowShoulder)
+            # x-y refers to the center --> offset to topLeft point
+            # handRectangle.x -= handRectangle.width / 2.f;
+            # handRectangle.y -= handRectangle.height / 2.f;
+            x -= width / 2
+            y -= width / 2  # width = height
+            # overflow the image
+            if x < 0: x = 0
+            if y < 0: y = 0
+            width1 = width
+            width2 = width
+            if x + width > image_width: width1 = image_width - x
+            if y + width > image_height: width2 = image_height - y
+            width = min(width1, width2)
+            # the max hand box value is 20 pixels
+            if width >= 20:
+                detect_result.append([int(x), int(y), int(width), is_left])
+
+    '''
+    return value: [[x, y, w, True if left hand else False]].
+    width=height since the network require squared input.
+    x, y is the coordinate of top left 
+    '''
+    return detect_result
+
+
+# Written by Lvmin
+def faceDetect(candidate, subset, oriImg):
+    # left right eye ear 14 15 16 17
+    detect_result = []
+    image_height, image_width = oriImg.shape[0:2]
+    for person in subset.astype(int):
+        has_head = person[0] > -1
+        if not has_head:
+            continue
+
+        has_left_eye = person[14] > -1
+        has_right_eye = person[15] > -1
+        has_left_ear = person[16] > -1
+        has_right_ear = person[17] > -1
+
+        if not (has_left_eye or has_right_eye or has_left_ear or has_right_ear):
+            continue
+
+        head, left_eye, right_eye, left_ear, right_ear = person[[0, 14, 15, 16, 17]]
+
+        width = 0.0
+        x0, y0 = candidate[head][:2]
+
+        if has_left_eye:
+            x1, y1 = candidate[left_eye][:2]
+            d = max(abs(x0 - x1), abs(y0 - y1))
+            width = max(width, d * 3.0)
+
+        if has_right_eye:
+            x1, y1 = candidate[right_eye][:2]
+            d = max(abs(x0 - x1), abs(y0 - y1))
+            width = max(width, d * 3.0)
+
+        if has_left_ear:
+            x1, y1 = candidate[left_ear][:2]
+            d = max(abs(x0 - x1), abs(y0 - y1))
+            width = max(width, d * 1.5)
+
+        if has_right_ear:
+            x1, y1 = candidate[right_ear][:2]
+            d = max(abs(x0 - x1), abs(y0 - y1))
+            width = max(width, d * 1.5)
+
+        x, y = x0, y0
+
+        x -= width
+        y -= width
+
+        if x < 0:
+            x = 0
+
+        if y < 0:
+            y = 0
+
+        width1 = width * 2
+        width2 = width * 2
+
+        if x + width > image_width:
+            width1 = image_width - x
+
+        if y + width > image_height:
+            width2 = image_height - y
+
+        width = min(width1, width2)
+
+        if width >= 20:
+            detect_result.append([int(x), int(y), int(width)])
+
+    return detect_result
+
+
+# get max index of 2d array
+def npmax(array):
+    arrayindex = array.argmax(1)
+    arrayvalue = array.max(1)
+    i = arrayvalue.argmax()
+    j = arrayindex[i]
+    return i, j

annotator/dwpose/wholebody.py

@@ -0,0 +1,49 @@
+import cv2
+import numpy as np
+
+import onnxruntime as ort
+from .onnxdet import inference_detector
+from .onnxpose import inference_pose
+
+class Wholebody:
+    def __init__(self):
+        device = 'cuda:0'
+        providers = ['CPUExecutionProvider'
+                 ] if device == 'cpu' else ['CUDAExecutionProvider']
+        # providers = ['CPUExecutionProvider']
+        providers = ['CUDAExecutionProvider']
+        onnx_det = 'annotator/ckpts/yolox_l.onnx'
+        onnx_pose = 'annotator/ckpts/dw-ll_ucoco_384.onnx'
+
+        self.session_det = ort.InferenceSession(path_or_bytes=onnx_det, providers=providers)
+        self.session_pose = ort.InferenceSession(path_or_bytes=onnx_pose, providers=providers)
+    def __call__(self, oriImg):
+        det_result = inference_detector(self.session_det, oriImg)
+        keypoints, scores = inference_pose(self.session_pose, det_result, oriImg)
+
+        keypoints_info = np.concatenate(
+            (keypoints, scores[..., None]), axis=-1)
+        # compute neck joint
+        neck = np.mean(keypoints_info[:, [5, 6]], axis=1)
+        # neck score when visualizing pred
+        neck[:, 2:4] = np.logical_and(
+            keypoints_info[:, 5, 2:4] > 0.3,
+            keypoints_info[:, 6, 2:4] > 0.3).astype(int)
+        new_keypoints_info = np.insert(
+            keypoints_info, 17, neck, axis=1)
+        mmpose_idx = [
+            17, 6, 8, 10, 7, 9, 12, 14, 16, 13, 15, 2, 1, 4, 3
+        ]
+        openpose_idx = [
+            1, 2, 3, 4, 6, 7, 8, 9, 10, 12, 13, 14, 15, 16, 17
+        ]
+        new_keypoints_info[:, openpose_idx] = \
+            new_keypoints_info[:, mmpose_idx]
+        keypoints_info = new_keypoints_info
+
+        keypoints, scores = keypoints_info[
+            ..., :2], keypoints_info[..., 2]
+        
+        return keypoints, scores
+
+

app.py

@@ -16,6 +16,7 @@ from kolors.models.unet_2d_condition import UNet2DConditionModel
 from diffusers import EulerDiscreteScheduler
 from PIL import Image
 from annotator.midas import MidasDetector
+from annotator.dwpose import DWposeDetector
 from annotator.util import resize_image, HWC3
 
 
@@ -24,6 +25,7 @@ ckpt_dir = snapshot_download(repo_id="Kwai-Kolors/Kolors")
 ckpt_dir_depth = snapshot_download(repo_id="Kwai-Kolors/Kolors-ControlNet-Depth")
 ckpt_dir_canny = snapshot_download(repo_id="Kwai-Kolors/Kolors-ControlNet-Canny")
 ckpt_dir_ipa = snapshot_download(repo_id="Kwai-Kolors/Kolors-IP-Adapter-Plus")
+ckpt_dir_pose = snapshot_download(repo_id="Kwai-Kolors/Kolors-ControlNet-Pose")
 
 text_encoder = ChatGLMModel.from_pretrained(f'{ckpt_dir}/text_encoder', torch_dtype=torch.float16).half().to(device)
 tokenizer = ChatGLMTokenizer.from_pretrained(f'{ckpt_dir}/text_encoder')
@@ -33,6 +35,7 @@ unet = UNet2DConditionModel.from_pretrained(f"{ckpt_dir}/unet", revision=None).h
 
 controlnet_depth = ControlNetModel.from_pretrained(f"{ckpt_dir_depth}", revision=None).half().to(device)
 controlnet_canny = ControlNetModel.from_pretrained(f"{ckpt_dir_canny}", revision=None).half().to(device)
+controlnet_pose = ControlNetModel.from_pretrained(f"{ckpt_dir_pose}", revision=None).half().to(device)
 
 image_encoder = CLIPVisionModelWithProjection.from_pretrained(f'{ckpt_dir_ipa}/image_encoder',  ignore_mismatched_sizes=True).to(dtype=torch.float16, device=device)
 ip_img_size = 336
@@ -62,12 +65,23 @@ pipe_canny = StableDiffusionXLControlNetImg2ImgPipeline(
     force_zeros_for_empty_prompt=False
 )
 
+pipe_pose = StableDiffusionXLControlNetImg2ImgPipeline(
+    vae=vae,
+    controlnet = controlnet_pose,
+    text_encoder=text_encoder,
+    tokenizer=tokenizer,
+    unet=unet,
+    scheduler=scheduler,
+    force_zeros_for_empty_prompt=False
+)
+
 for pipe in [pipe_depth]:
     if hasattr(pipe.unet, 'encoder_hid_proj'):
         pipe.unet.text_encoder_hid_proj = pipe.unet.encoder_hid_proj
         
 pipe_depth.load_ip_adapter(f'{ckpt_dir_ipa}' , subfolder="", weight_name=["ip_adapter_plus_general.bin"])
 pipe_canny.load_ip_adapter(f'{ckpt_dir_ipa}' , subfolder="", weight_name=["ip_adapter_plus_general.bin"])
+pipe_pose.load_ip_adapter(f'{ckpt_dir_ipa}' , subfolder="", weight_name=["ip_adapter_plus_general.bin"])
 
 @spaces.GPU
 def process_canny_condition(image, canny_threods=[100,200]):
@@ -88,6 +102,16 @@ def process_depth_condition_midas(img, res = 1024):
     result = cv2.resize(result, (w,h))
     return Image.fromarray(result)
 
+model_dwpose = DWposeDetector()
+@spaces.GPU
+def process_dwpose_condition(image, res=1024):
+    h,w,_ = image.shape
+    img = resize_image(HWC3(image), res)
+    out_res, out_img = model_dwpose(image) 
+    result = HWC3(out_img)
+    result = cv2.resize( result, (w,h) )
+    return Image.fromarray(result)
+
 MAX_SEED = np.iinfo(np.int32).max
 MAX_IMAGE_SIZE = 1024
 
@@ -165,6 +189,39 @@ def infer_canny(prompt,
     ).images[0]
     return [condi_img, image], seed
 
+@spaces.GPU
+def infer_pose(prompt, 
+          image = None, 
+          negative_prompt = "nsfw，脸部阴影，低分辨率，jpeg伪影、模糊、糟糕，黑脸，霓虹灯", 
+          seed = 66, 
+          randomize_seed = False,
+          guidance_scale = 6.0, 
+          num_inference_steps = 50,
+          controlnet_conditioning_scale = 0.7,
+          control_guidance_end = 0.9,
+          strength = 1.0
+        ):
+    if randomize_seed:
+        seed = random.randint(0, MAX_SEED)
+    generator = torch.Generator().manual_seed(seed)
+    init_image = resize_image(image,  MAX_IMAGE_SIZE)
+    pipe = pipe_pose.to("cuda")
+    condi_img = process_dwpose_condition(np.array(init_image), MAX_IMAGE_SIZE)
+    image = pipe(
+        prompt= prompt ,
+        image = init_image,
+        controlnet_conditioning_scale = controlnet_conditioning_scale,
+        control_guidance_end = control_guidance_end, 
+        strength= strength , 
+        control_image = condi_img,
+        negative_prompt= negative_prompt , 
+        num_inference_steps= num_inference_steps, 
+        guidance_scale= guidance_scale,
+        num_images_per_prompt=1,
+        generator=generator,
+    ).images[0]
+    return [condi_img, image], seed
+
 canny_examples = [
     ["一个红色头发的女孩，唯美风景，清新明亮，斑驳的光影，最好的质量，超细节，8K画质",
      "image/woman_2.png", "image/2.png"],
@@ -175,6 +232,12 @@ depth_examples = [
      "image/1.png","image/woman_1.png"],
 ]
 
+pose_examples = [
+    ["一位穿着紫色泡泡袖连衣裙、戴着皇冠和白色蕾丝手套的女孩，超高分辨率，最佳品质，8k画质",
+     "image/woman_3.png","image/woman_4.png"],
+]
+
+
 css="""
 #col-left {
     margin: 0 auto;
@@ -270,6 +333,7 @@ with gr.Blocks(css=css) as Kolors:
             with gr.Row():
                 canny_button = gr.Button("Canny", elem_id="button")
                 depth_button = gr.Button("Depth", elem_id="button")
+                pose_button = gr.Button("Pose", elem_id="button")
             
         with gr.Column(elem_id="col-right"):
             result = gr.Gallery(label="Result", show_label=False, columns=2)
@@ -291,6 +355,14 @@ with gr.Blocks(css=css) as Kolors:
                 outputs = [result, seed_used],
                 label = "Depth"
             )
+    with gr.Row():
+        gr.Examples(
+                fn = infer_pose,
+                examples = pose_examples,
+                inputs = [prompt, image, ipa_image],
+                outputs = [result, seed_used],
+                label = "Pose"
+            )
 
     canny_button.click(
         fn = infer_canny,
@@ -304,4 +376,10 @@ with gr.Blocks(css=css) as Kolors:
         outputs = [result, seed_used]
     )
 
+    pose_button.click(
+        fn = infer_pose,
+        inputs = [prompt, image, ipa_image, negative_prompt, seed, randomize_seed, guidance_scale, num_inference_steps, controlnet_conditioning_scale, control_guidance_end, strength, ip_scale],
+        outputs = [result, seed_used]
+    )
+
 Kolors.queue().launch(debug=True)

requirements.txt

@@ -7,4 +7,5 @@ sentencepiece==0.1.99
 torchvision==0.17.0
 timm
 einops
-scipy
+scipy
+onnxruntime-gpu