WindVChen

Hi @BrotherQi , Could you please share more details about this issue? We've successfully run the `efficient_inference_for_square_image.py` code in our tests.

Hi @D-Mad , I've thoroughly tested the code on an alternate device and have indeed identified the peculiar phenomenon you mentioned. It's quite perplexing, given that the code runs seamlessly...

Hi @wasahaiah, You can raise an issue under [HRNetv2 repository](https://github.com/HRNet/HRNet-Image-Classification) for this problem. We only provide their download link in our README.

Hi @wzking09 , To my knowledge, currently, the quite efficient way is to predict the params of color filters, which can lead to a lightweight and efficient network easily (yet...

Hi @Willoverseas , May I know what error you encountered?

Hi @TLDX-XIONG , Yes, I did manage to solve the problem. However, it's been a while since I worked on it, so I don't recall all the specifics of the...

Hi @raniasyed , I didn't encounter such an issue before, but maybe you can have a look at this issue https://github.com/huggingface/transformers/issues/17611 which is quite similar. Besides, there could also sometimes...

Hi @raniasyed , The time cost mainly comes from multiple iterations of text embeddings optimization (code refer to [here](https://github.com/WindVChen/Diff-Harmonization/blob/7d469e698924c27bf9f8e35d12c37296185e30b7/diff_harmon.py#L474)), null-text embeddings optimization (code refer to [here](https://github.com/WindVChen/Diff-Harmonization/blob/7d469e698924c27bf9f8e35d12c37296185e30b7/diff_harmon.py#L484)), and the multiple rounds...

哈喽 @Alexwong-1024 , 都是在测试集测试的哈.

文章中的是500，实际要训得话也可以调到1000，两者性能不会有太大差异，但1000步收敛更稳定些~