Gaurav Parmar
Gaurav Parmar
We use FID and structure similarity to evaluate the unpaired models.
This typically happens when the InceptionV3 file did not get downloaded completely. Can you run this command to check if you get any error? ``` from cleanfid.fid import build_feature_extractor feat_model...
Oh I apologize, I had a typo in the above command. @YijiFeng @and-human Could you try this command below instead to see if you downloaded the `clean-fid` weights correctly: ```...
Thank you for the interest in this project! The files `fixed_prompt_a.txt` and `fixed_prompt_b.txt` should contain a single caption that describe the domains A and B respectively. For instance for horse2zebra...
Great point! If you end up trying it and training a model, I would be happy to include that in this repo! -Gaurav
We use the BDD100k driving dataset for the day to night conversion model. For dataset processing, we resize all images to 512x512 during training and inference. The other configurations should...
>"driving in the night" for a2b and "driving in the day" for b2a? Yup that is correct.
Hi, Thanks for your interest in this project! Yeah, we will release the model weights for rainy2clear and night2day as well! -Gaurav
Hopefully in a few days, as soon as I clean up the relevant code!
The pretrained model and the corresponding code for using them for other unpaired tasks (night2day, clear2rainy, rainy2clear) are uploaded. See the instructions [here](https://github.com/GaParmar/img2img-turbo/tree/main?tab=readme-ov-file#getting-started). @scarbain