EnYu comments

Results 6 comments of


                                            EnYu

youtube_vos: No test meta_expressions.json

Thanks a lot. I also have another problem. There will be CUDA out of memory when I inference ytvos, however I can train the model in ytvos normally. Is there...

about training cost

Hi, Merlin is the pretrained weights and Merlin-Chat is the weights after SFT. The entire training process is conducted on 64 NVIDIA A800 GPUs, with approximately 12 hours required for...

how to launch local inference

We apologize for the time constraints; we have not yet organized the code to support multi-round, multi-frame video demos. However, at this stage, we support single-round dialogues, and you can...

Release dataset

Hi, Thank you for your attention to our work. We will open-source the Merlin-Chat SFT data after ECCV. Stay tuned for further updates.

To use vicuna model as text decoder

Thanks for your attention. Yeah, you need to download the whole files of vicuna-v15. CUDA version is cuda 11.8.

How can I get the clip-vit-large-patch14-448

Thanks for your attention. There is no original clip-vit-large-patch14-448 on the hugging face. We employed a positional embedding interpolation to adapt the original 224x clip-vit to support an input resolution...