Closed john-backlot closed 1 year ago
Hi @john-backlot Thanks for your interest in our work!!
It appears that the code is running in the 'paired' setting, where the in-shop garment matches the one worn by the model. Consequently, the resulting image closely resembles the input image.
To obtain the desired results, I recommend running the inference with the flag --test_order unpaired. In this setting, the in-shop garment differs from what the model is wearing. By using this flag, you should be able to achieve the expected outcomes.
Alberto
We're running inference to recreate paper results using the VITON-HD dataset (_testpairs.txt in our conda environment), however the results appear to be slightly modified versions of the original clothing. It appears as if it's going through a diffusion pass, but not applying the clothing.
Running on a Windows 11 4090 PC, following the default settings/commands provided.
Images:
![final_image](https://github.com/miccunifi/ladi-vton/assets/79934252/098a5b9e-aefa-47ea-893d-869f97f78d81)