search

Tangshitao / MVDiffusion

MVDiffusion: Enabling Holistic Multi-view Image Generation with Correspondence-Aware Diffusion, NeurIPS 2023 (spotlight)
447 stars 21 forks source link

About Demo test #33

Open Mr-Ma-yikun opened 8 months ago

Mr-Ma-yikun commented 8 months ago

When performing a demo test for Dual Conditioned Generation, I found that the inpainting images cannot be generated properly. Only the mask is being outputted, with no inpainting content.

Tangshitao commented 8 months ago

Can you post 8 generated images here?

Mr-Ma-yikun commented 8 months ago

0 1 2 3 4 5 6 7 pano

Tangshitao commented 8 months ago

Have you modify the codes. I feel the image preprocess is not correct

Mr-Ma-yikun commented 8 months ago

Have you modify the codes. I feel the image preprocess is not correct

I have not made any modifications to the code. The only additions I made were pano_outpaint.ckpt and stable-diffusion-2-inpainting.

Tangshitao commented 8 months ago

Can you check try diffusers 0.16.1 and make sure all the weights are load correctly?