Pretrained renet models

[EricYang3721]

Hello @JinhwiPark,

Thanks for releasing the code. When I run the code it needs to load the pretained ResNet models as model_path = { 'resnet18': 'pretrained/COMMON/resnet18.pth', 'resnet34': 'pretrained/COMMON/resnet34.pth' } I didn't find these models in the url of your pre-trained weights. Could you help to release those or point us to the place to download those models, please?

Thank you very much!

[EricYang3721]

I wasn't able to reproduce the NYUv2 performance with downloaded models. Could you help to point out me where I got wrong, please?

1. Without using resnet model

   python test_multiDataLoader.py --data_name NYU --dir_data /home/nrasrtf/src/src/DepthPrompting/data/nyudepth_hdf5  --gpus 0,1,2,3 --model_name depth_prompt_main --pretrain /home/nrasrtf/src/src/DepthPrompting/pretrained_weights/Depthprompting_depthformer_kitti.tar --prop_kernel 9 --conf_prop --prop_time 18 --patch_height 240 --patch_width 320 --nyu_val_samples 500,200,100,5,1 --init_scailing   --no_res_pre

   Got following results

   Test for various Sampels/Lidars: [500, 200, 100, 5, 1]
   0.3117/0.2065 0.3416/0.2289 0.3839/0.2619 4.4319/4.1520 4.8406/4.6754

2. Downloaded resent base model from pytorch official website https://pytorch.org/vision/0.8/_modules/torchvision/models/resnet.html, and run following

   python test_multiDataLoader.py --data_name NYU --dir_data /home/nrasrtf/src/src/DepthPrompting/data/nyudepth_hdf5  --gpus 0,1,2,3 --model_name depth_prompt_main --pretrain /home/nrasrtf/src/src/DepthPrompting/pretrained/OURS/Depthprompting_depthformer_kitti.tar --prop_kernel 9 --conf_prop --prop_time 18 --patch_height 240 --patch_width 320 --nyu_val_samples 500 --init_scailing

   The model Depthprompting_depthformer_kitti.tar is from the official URL provided, didn't do any fine tuning or training.

   Test for various Sampels/Lidars: [500]
   0.3117/0.2065