Konstantin

@walzimmer Have you succeeded in solving this? AFAIU there are no separate cameras intrinsic parameters setup in the architecture so the network is trained to work with specific camera only....

I've managed to run pretrain_stage_2 on a single 3090 (opt-2.7b, batch size = 32, single worker), train_caption_coco crashes with OOM though. I've noticed that image size is different for these...