slowfast_detection does not work with the demo code in tutorial

[botamochi6277]

❓ Questions on how to use slowfast_r50_detection model

I already succeed to run slow_r50_detection model with the code in detection tutorial.
After it, I tried to run slowfast_r50_detection with change parameters and failed.

I just changed following items

• change the model class: from slow_r50_detection to slowfast_r50_detection
• change parameters for preprocessing : from inputs, inp_boxes, _ = ava_inference_transform(inp_imgs, predicted_boxes.numpy()) to inputs, inp_boxes, _ = ava_inference_transform(inp_imgs, predicted_boxes.numpy(), num_frames=32, slow_fast_alpha=4)

I got the following error:

...
preds = video_model(inputs.unsqueeze(0).to(device), inp_boxes.to(device))
                      ^^^^^^^^^^^^^
AttributeError: 'list' object has no attribute 'unsqueeze'

---

This inputs is clip output of ava_inference_transform. Please look outputs of ava_inference_transform function. Its 'clip' output is torch.Tensor when slow_fast_alpha is None. On the other hand, 'clip' is list[torch.Tensor] when slow_fast_alpha is 4.

Do you have any idea about the way converting clip to suitable format for slowfast_r50_detection?