Yuxuan Zhang

Must I use regulization imgs retrieved from laion?

2

Thanks for your wonderful work! When i want to train a model for myself, i found that the code starts to download imgs from laion. Is this process necessary? Could...

Hi, appreciate your nice work. Observed that there is no layernorm in text_hidden_fcs module and dropout rate is set to 0. Are there any experiments that support such design?

Confused about which one to use while evaluating the model.

Is it necessary decoding bbox regression before fam-3d?

1

Hi, congradulation on your works! Trying to add some modules based on your design, I found that at [https://github.com/Adamdad/ConsistentTeacher/blob/main/ssod/models/dense_heads/fam3d.py#L168](url), the bbox regression is decoded before fam-3d. Is this necessary? May...

Is there an update for adapter.py?

3

Thanks for your pretraining! I want to use your T2I-canny pretrained ckpt, but the current adapter.py seems mismatch with the ckpt >The config attributes {'name': 'canny'} were passed to T2IAdapter,...

Why warning "A parameter name that contains `gamma` will be renamed internally to `weight`."

1

As mentioned in [SAM-2 Issue 85](https://github.com/facebookresearch/segment-anything-2/issues/85), there are modules named "gamma" in [SAM-2's memory encoder](https://github.com/facebookresearch/segment-anything-2/blob/57bc94b7391e47e5968004a0698f8bf793a544d1/sam2/modeling/memory_encoder.py#L97), which is conflict with [transformers](https://github.com/huggingface/transformers/blob/main/src/transformers/modeling_utils.py#L698). (transformers forbid module names "gamma" or "beta") To solve this...