Has anyone tried training AI images generation models(e.g. lora for stable diffusion) with heic images read with pillow-heif? Resulting generated images look weird, aspect ratio is somehow corrupted and people's head tends to get wider. Also, it affects learning as well, the model underfits. What are some potential points that I might be missing?
P.S When I check the image before feeding it to stable diffusion, everything looks normal.
Edit: It was not related to piilow-heif library, issue can be closed.
Has anyone tried training AI images generation models(e.g. lora for stable diffusion) with heic images read with pillow-heif? Resulting generated images look weird, aspect ratio is somehow corrupted and people's head tends to get wider. Also, it affects learning as well, the model underfits. What are some potential points that I might be missing?
P.S When I check the image before feeding it to stable diffusion, everything looks normal.
Edit: It was not related to piilow-heif library, issue can be closed.