Metric Depth Estimation is all black

[MichaelWangGo]

Hi @LiheYoung ,

I used script python train_mono.py -m zoedepth -d scared --pretrained_resource="" to train the Metric Depth Estimation model on my own dataset (named scared), and got the model which was stored in './depth_anything_finetune', but when I inferenced the model using python evaluate.py -m zoedepth --pretrained_resource="local::./depth_anything_finetune/ZoeDepthv1_06-Feb_11-00-1eb3234d84fb_latest.pt" -d scared_test, all the predictions are black. I tried many times and got the same results, can you give me some suggestions?

I added information of the my own datasets in the ./zoedepth/utils/config.py: ` "scared": { "dataset": "scared", "min_depth": 0.001, "max_depth": 25, "data_path": os.path.join(Scared_HOME_DIR, "keyframe_scared_train"), "gt_path": os.path.join(Scared_HOME_DIR, "keyframe_scared_train"), "filenames_file": "./train_test_inputs/scared_train_files_with_gt_copy.txt", "input_height": 1024, "input_width": 1280, # 704 "data_path_eval": os.path.join(Scared_HOME_DIR, "keyframe_scared_val"), "gt_path_eval": os.path.join(Scared_HOME_DIR, "keyframe_scared_val"), "filenames_file_eval": "./train_test_inputs/scared_val_files_with_gt_copy.txt",

    "min_depth_eval": 0.001,
    "max_depth_eval": 25,

    "do_random_rotate": True,
    "degree": 1.0,
    "do_kb_crop": True,
    "garg_crop": True,
    "eigen_crop": False,
    "use_right": False
},
    "scared_test": {
    "dataset": "scared",
    "min_depth": 0.001,
    "max_depth": 25,
    # "data_path": os.path.join(Scared_HOME_DIR, "keyframe_scared_train"),
    # "gt_path": os.path.join(Scared_HOME_DIR, "keyframe_scared_train"),
    # "filenames_file": "./train_test_inputs/scared_train_files_with_gt.txt",
    "input_height": 1024,
    "input_width": 1280,  # 704
    "data_path_eval": os.path.join(Scared_HOME_DIR, "keyframe_scared_test"),
    "gt_path_eval": os.path.join(Scared_HOME_DIR, "keyframe_scared_test"),
    "filenames_file_eval": "./train_test_inputs/scared_test_files_with_gt.txt",
    "save_images": "./scared_test",
    "min_depth_eval": 0.001,
    "max_depth_eval": 25,

    "do_random_rotate": True,
    "degree": 1.0,
    "do_kb_crop": True,
    "garg_crop": True,
    "eigen_crop": False,
    "use_right": False
},`

[LiheYoung]

Hi, did you successfully train your model with your data?

[MichaelWangGo]

When I finish training, I got the result here: