fix(demo): trailing comma wraps bbox array in a tuple (+2 more)

[andrewwhitecdw]

Small fixes in pose_est/hybrik_demo/demo.py:

fix: trailing comma wraps bbox array in a tuple

Fix: Replace:

        mot_bboxes[idx]['bbox'] = np.stack(bboxes[idx]),

with:

        mot_bboxes[idx]['bbox'] = np.stack(bboxes[idx])

fix: single-person branch skips frames and mismatches bbox/exist lengths

Fix: Apply patch:

--- a/pose_est/hybrik_demo/demo.py
+++ b/pose_est/hybrik_demo/demo.py
@@ -226,10 +226,11 @@
     prev_box = None
 
     out_dict = defaultdict(lambda: defaultdict(list))
     idx = 0   # single person id
 
-    frame_idx = 0
+    frame_idx = -1
 
     bbox_exist = []
-    bboxes = []
+    bboxes = [np.zeros(4) for _ in range(len(img_path_list))]
     for img_path in tqdm(img_path_list):
+        frame_idx += 1
         dirname = os.path.dirname(img_path)
         basename = os.path.basename(img_path)
 
@@ -294,7 +295,7 @@
         pts[:, 0] = pts[:, 0] + bbox_xywh[0]
         pts[:, 1] = pts[:, 1] + bbox_xywh[1]
 
-        bboxes.append(np.array(bbox_xywh))
+        bboxes[frame_idx] = np.array(bbox_xywh)
 
         new_princpt = np.array([image.shape[1], image.shape[0]]) * 0.5
         transl[:2] += (np.array(princpt) - new_princpt) * transl[2] / np.array(focal)
@@ -309,7 +310,6 @@
         out_dict[idx]['cam_K'].append(K.astype(np.float32))
 
         image_vis = cv2.cvtColor(image_vis, cv2.COLOR_RGB2BGR)
-        frame_idx += 1
         res_path = os.path.join(opt.out_dir, 'res_images', f'{frame_idx:06d}.jpg')
         cv2.imwrite(res_path, image_vis)

fix: multi-person bbox list length mismatches exist array

Fix: Apply patch:

--- a/pose_est/hybrik_demo/demo.py
+++ b/pose_est/hybrik_demo/demo.py
@@ -123,6 +123,7 @@
     # initialize
     for person_id in tracking_results:
         bbox_exist[person_id-1] = [0 for _ in range(len(img_path_list))]
+        bboxes[person_id-1] = [np.zeros(4) for _ in range(len(img_path_list))]
 
     for frame_idx in tqdm(range(len(img_path_list))):
 
@@ -169,7 +170,7 @@
                 pts[:, 0] = pts[:, 0] + bbox_xywh[0]
                 pts[:, 1] = pts[:, 1] + bbox_xywh[1]
 
-                bboxes[idx].append(np.array(bbox_xywh))
+                bboxes[idx][frame_idx] = np.array(bbox_xywh)
 
                 new_princpt = np.array([image.shape[1], image.shape[0]]) * 0.5
                 transl[:2] += (np.array(princpt) - new_princpt) * transl[2] / np.array(focal)

Files changed

• pose_est/hybrik_demo/demo.py