Closed johangutihaz closed 2 years ago
Hi, 1) Hi, we do the search phase and training phase on SceneFlow dataset by NVIDA V100 GPU. For Middlebury, as we take the input_size of half of the original images, it needs a GPU ( RTX8000 48GB ) with a larger memory. 2) Please check the best models on each benchmark.
What are the requirements for the GPU? In the paper the NVIDA V100 GPU is mentioned for the trainingstep but here in the comments RTX8000 48GB and GeForce 3090 are mentioned.
Also, is there any trained model for use?