Long(Tony) Lian
Long(Tony) Lian
Can you try contacting the authors of DenseCL to see if they can fix this? The link is hosted by them. The link is from this page: https://github.com/aim-uofa/AdelaiDet/blob/master/configs/DenseCL/README.md
USL-T implementation is a little different to USL, as it requires training while this codebase currently doesn't support training. I'm planning to integrate it into this codebase. Does giving you...
I added the indices in README.
Added USL-T implementation, checkpoints, and sample selection. Feel free to contact me if you have any questions.
Working on the camera ready paper. We will release our code at about the same time.
We are still organizing the code repo, which will include both the LLM text-to-DSL part and DSL-to-video part based on cross-attention control. However, similar to LMD+, we offer a simple...
@jianlong-yuan @Bailey-24 @guyuchao @blE-lj Thanks for your interest! The code and the benchmark have been released. Feel free to ask any questions.
The problem is likely due to `_ignore_causal_mask_sdpa` returning `False` (called by `_prepare_4d_causal_attention_mask_for_sdpa`), **when the sequence is longer than [the sliding window size](https://huggingface.co/google/gemma-2-9b-it/blob/main/config.json#L28) 4096**, causing `attention_mask` to be not None. The...
LVD is training-free. If you're asking about IGLIGEN, 4x A6000 will work.
If you set `lvd_gligen_scheduled_sampling_beta` to 0, it is equivalent to not using the GLIGEN adapters trained in IGLIGEN project. This way it falls back to the original baseline model (since...