Geonmo Gu

Hi guys. I tested NLLB using huggingface transformers. NOTE: You should install the latest dev version using below instruction in order to use NLLB tokenizer. ``` $ pip install git+https://github.com/huggingface/transformers.git...

Hi Hinnibal7ecter! simple approach: 1. Slice your large image to several available size images like 720p or less 2. Enhance sliced images separately 3. Concatenate enhanced sliced images into one...

Hannibal7ecter, Please comment in English. It's hard to understand your questions thr translator. :cry: simple approach: + Run Enhancenet using CPU, not GPU. + Its free from 8GB GPU mem,...

Sorry for the late response. I added the license file (MIT license). Feel free to use it!

issue 남겨 주셔서 감사합니다. 학습 과정 중에 발생된 error에 대해 gradient가 모든 layer를 거치면서 neural networks가 이를 잘 대응하고 있는지 확인하기 위함입니다. 학습이 잘 되지 않는 경우 이런식으로 확인하곤 하는데요....

해석은 weight 값의 변화가 있냐 없냐로 단순하게 생각하시면 될 것 같아요.  위 식에 의해 weight들이 loss (L)를 감소하는 방향 (-) 으로 학습이 진행됩니다. 표현은 단순하나 그 속엔 쌓아 놓은...

Try this before Composer comes out. It's Graphit we released and you might find something interesting. I also hope Composer will be released soon. https://github.com/navervision/Graphit

Hi, jhpinkorea. Negative pairs are nested in anchor_features*pos_features. The main diagonal elements in anchor_features*pos_features are positive pairs and others are negative pairs.

Hi there. https://github.com/navervision/CompoDiff/blob/50e5ff6a60d60ffcc5ea47b751989baba9ed2ee3/demo_search.py#L119 In the above line, you can get the **composed** image features. https://github.com/navervision/CompoDiff/blob/50e5ff6a60d60ffcc5ea47b751989baba9ed2ee3/demo_search.py#L133-L147 and here you can retrieve top-K image urls using the composed image features. Can you...

Yes sure. This demo doesn't have the evaluate protocol for benchmark datasets. We have currently only released the demo code to show the generalization performance of CompoDiff.