Possible Issue with `bboxes_batch` Initialization in `lite/demo/vis_pose.py`

[FrankWuuu]

---

Hi,

I've been working with the lite version of the pose estimation demo, specifically the file lite/demo/vis_pose.py. I noticed a potential issue with the initialization of bboxes_batch in lines 394 to 398. The current code is as follows:

for i, bboxes in enumerate(bboxes_batch):
    if len(bboxes) == 0:
        bboxes_batch[i] = np.array(
            [[0, 0, orig_img_shape[1], orig_img_shape[0]]]
        )

It seems like the last element of the array should use orig_img_shape[2] instead of orig_img_shape[0]. Therefore, the corrected version might look like this:

for i, bboxes in enumerate(bboxes_batch):
    if len(bboxes) == 0:
        bboxes_batch[i] = np.array(
            [[0, 0, orig_img_shape[1], orig_img_shape[2]]]
        )

This change aligns with the typical (height, width) order, cuz orig_img_shape follows the format (B, H, W, C).

Could you please verify whether this adjustment is necessary? Thank you for your attention to this detail!

Best regards, Frank Wu

[rawalkhirodkar]

@FrankWuuu thank you for spotting this. When the detection failed, this would result in incorrect bbox size. Your PR is merged.