Lightnessly

awesome work!!!!!!

3

很好的repo，帮助我这个菜鸟快速了解nlp分词领域，请问一下可以提供训练之后的model嘛，感谢

我好奇Nextvlad在其他数据集上的表现，但是网上只有youtube的，请问有哪位大神比较过nexvlad和TSM在相同数据集上的表现嘛 感谢

Hi, thanks for your awesome work. when I use fvcore to count flops I find some ops are not supported. when I count inference time between TAM and Tada R50....

Hi, I found the head conv channels is 64 but the ori paper is 24. Is there any reason to change the channels?

根据两者官网和自己复现，性能基本属实，请问可以公开是如何训练的嘛？如果用的也是公开数据集，就打破了我对跨模态pretrain的模型大小和性能关系的认知。还是说ERNIE用的是 fine-tune的结果。因为我看Wukong的finetune结果和ERNIE的比较接近

Can not get fine result when type is real_occl

1

Thanks for your re-production. When I test reall image with occl type, it seems that the model is overfitted with irregular mask，and predicts the wrong mask with irregular lines like...

Salute your open source spirit！

1

The earth and the sky will praise your generosity, and countless algorithm engineers will praise you, the selfless devotee, the great architect.

[SD3] vae.config.shift_factor missing in dreambooth training examples

6

### Describe the bug shift_factor missing in traning code: https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth_lora_sd3.py#L1617, but used in inference code: https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py#L893 Is it resonable that when traning SD3, we do not need to norm latents...

what prompt format shoud be if I want to input an image and a video at the same time

1

Thanks for opening source the powfer model. the case show how to generate prompt for text-image and video-image. but If I want to input image-video-text at the same time, how...

Your work is very helpful. The earth and the sky will praise your generosity, and countless algorithm engineers will praise you, the selfless devotee, the great architect.