Kephone Gaon

some questions about quantization and shift

1

Thank you for your amazing work,. I have carefully read it but still have some questions. (1) in section 3.3.2, I understand " batch normalization degenerates into to a scaling...

https://github.com/Scalsol/mega.pytorch/blob/a6aa6e0537b82d70da94228100a51e6a53d98f82/mega_core/data/datasets/vid_mega.py#L133 `images["cur"] = img`, However, this `img` is not corresponding to the current `idx` when `cfg.MODEL.VID.MEGA.GLOBAL.ENABLE == True`. The for loop in line116 will replace `img` at each iteration https://github.com/Scalsol/mega.pytorch/blob/a6aa6e0537b82d70da94228100a51e6a53d98f82/mega_core/data/datasets/vid_mega.py#L116...

Why is there `mega_core.trainer ERROR`?

5

``mega_core.trainer ERROR: Iteration=4 || Image names used for training ('train/ILSVRC2015_VID_train_0003/ILSVRC2015_train_00839000/000255',) || targets Length=[0]`` (I have modified the error log infor to print the image file name) In `self.filter_annotation()`, `class VIDMEGADataset`,...

Problems about The process to complete the separated areas

1

Tank you for your amazing work, I have some trouble to understand the process to complete the separated areas, i.e., Fig.4. in the paper. Can you point out which part...

classeme features

4

refer to https://github.com/Dawn-LX/VidSGG-BIG/issues/2#issuecomment-1122197607

questions about `args.use_matching`

5

Thank you for the nice work of HOI. I'm currently following your work I have a question about the args.use_matching. It seems that you did not use the `matching_embed` when...

Problem about the loss function

2

I am confused about the loss function of entity and predicate. In your paper, Eq.(10) and Eq.(11), you use the l2_loss between E^O and E^T **However**, in the code, it...

Problem about message passing process

2

I have some problem about your code at https://github.com/alirezazareian/vspnet/blob/68db0265b49107e0aa66430eea4cf56faa85aece/src/tensorflow/models/vsp_model_002.py#L372 Why update message of entity `if step > 0 ` while updating the message of predicate not require step > 0...