CUDE out of memory

[Haruko386]

Hi, excellent work! I'm tring to reproduce this paper on my own PC, which have a single RTX 4090 24GB, my training config is:

{
    "train_data_path" : "/root/Dataset/Hypersim/precessed/train",
    "val_data_path" : "/root/Dataset/Hypersim/precessed/val",
    "train_filenames_path" : "/root/EcoDepth/filenames/hypersim/filename_list_train_filtered.txt",
    "val_filenames_path" : "/root/EcoDepth/filenames/hypersim/filename_list_val_filtered_small_80.txt",
    "train_depth_factor" : 1000.0,
    "val_depth_factor" : 1000.0,
    "use_cut_depth" : false,
    "do_random_crop" : false,
    "crop_h" : 480,
    "crop_w" : 640,
    "batch_size" : 1,
    "num_workers" : 1,
    "max_depth" : 10.0,
    "no_of_classes" : 100,
    "train_from_scratch" : false,
    "eval_crop" : "eigen",
    "ckpt_path" : "/root/weights_indoor.ckpt",
    "scene" : "indoor",
    "epochs" : 25,
    "weight_decay" : 0.1,
    "lr" : 1e-5,
    "val_check_interval" : 1684
}

but still:

torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 4.50 GiB. GPU 0 has a total capacity of 23.64 GiB of which 361.19 MiB is free. Process 4160709 has 23.09 GiB memory in use. Of the allocated memory 14.75 GiB is allocated by PyTorch, and 7.87 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
Epoch 0:   0%|          | 0/53885 [00:02<?, ?it/s] 

Is that possible to train EcoDpeth on low memory card ?