code for inference on custom images

[francescobaraldi]

is there the code to run the model on custom images or video?

[yangjie-cv]

Thanks for your interest. We have uploaded the inference scripts for single-image virtualization. For the videos, you need to do inference frame by frame.

[julkaztwittera]

Could you please explain how does this script work? I have put an image in Inference_Path and annotations from Coco2017 dataset to EDPOSE_COCO_PATH. I tried to run the script Virtualization via COCO Keypoints Format, but I got the following error:

  File "/home/user/ED-Pose/util/misc.py", line 255, in log_every
    header, total_time_str, total_time / len(iterable)))
ZeroDivisionError: float division by zero

and the process hangs infinitely.

[yangjie-cv]

You need use the command: export Inference_Path=/path/to/your/inference_dir

[julkaztwittera]

I have already done it. I also downloaded COCO annotations and put them in EDPOSE_COCO_PATH, but it doesn't work.

Why does inference code on one image need the whole COCO dataset...?

[julkaztwittera]

Well, is there a simple way to just load the model and infer on a single image without using the dataloader? I tried to do this on a torch tensor, but there were some errors.