OOM and datasets issue

[wuyou012]

Awesome Work! Great to see such depth and camera position solutions! Unfortunately, after finding some solutions regarding the previous issues, I still struggle with OOM problem. CUDA_VISIBLE_DEVICES=0,1 python3 -m flowmap.overfit dataset=images dataset.images.root=/home/flowmap/data/tandt_caterpillar +experiment=low_memory I have successed using single card, and could you explain why we can't use mutiple cards. Also, could you explain the difference between the images and COLMAP dataset? I also would like to know the details about figure S4 in support materials, how could I refine such low resolution depth map?