search

hisfog / SfMNeXt-Impl

[AAAI 2024] Official implementation of "SQLdepth: Generalizable Self-Supervised Fine-Structured Monocular Depth Estimation", and more.
MIT License
91 stars 12 forks source link

I got bad results when reproducing #16

Open parallelarc opened 1 year ago

parallelarc commented 1 year ago

Nice work! But I got some problems when reproducing. This is my config:
--data_path ../raw_data/ --log_dir ./logdir/ --model_name res_640x192 --eval_split eigen --backbone resnet_lite --height 192 --width 640 --batch_size 16 --num_epochs 25 --scheduler_step_size 15 --num_layers 50 --num_features 256 --model_dim 32 --patch_size 16 --dim_out 64 --query_nums 64 --eval_mono --post_process --load_pretrained_model --load_pt_folder ./pretrained/ --pretrained_pose --pose_net_path ./pretrained/

I downloaded pretrained models (depth.pth, encoder.pth and pose.pth) in the "pretrained" folder and I used jpeg images. But I got some bad results like these. Could you help me...

1 2

hisfog commented 1 year ago

It seems that you did not normalize the depth map, which means you saved the raw depth values. For visualization, you should normalize the depth map to [0, 1] and then save it as an image.

parallelarc commented 1 year ago

It seems that you did not normalize the depth map, which means you saved the raw depth values. For visualization, you should normalize the depth map to [0, 1] and then save it as an image.

Thanks! I will have a try.