How to predict or inference on a single figure when performing depth estimation?

[QinganZhang]

I want to predict or inference on some figures instead of evaluting on NYUv2 to perform depth estimation, how should I do? I have some figures and I just hope to have a look of the performance on your model or method.

I just have a try, but I get confused when handling my own dataset(only test on dozens of figures) and meet some error.

So I wonder whether there is a convenient way or modifying least codes to do inference on a single figure? Thanks a lot.

[adam-kosinski]

@QinganZhang I was able to get it to run on my own images by running one of the evaluation commands, and changing a few options, based on the GLPDepth documentation (inference section), since this test script is adapted from GLPDepth.

• change --dataset to imagepath
• change --data_path to the directory with your images, make sure your images end in lowercase .jpg or .png
• add the --save_visualize option, saves visualizations in the results/ directory
• remove the --do_evaluate option, I think it caused an error because my images didn't have depths associated

Remember to download the ckpt file (or train to get it I guess?) and put it in the ckpt directory. Also see #1 if you're having errors about multiple devices.

For example:

python3 test.py --dataset imagepath --data_path my_images --save_visualize --max_depth 10.0 --max_depth_eval 10.0  --backbone swin_base_v2 --depths 2 2 18 2 --num_filters 32 32 32 --deconv_kernels 2 2 2 --window_size 30 30 30 15 --pretrain_window_size 12 12 12 6 --use_shift True True False False --flip_test --shift_window_test --shift_size 2 --ckpt_dir ckpt/nyudepthv2_swin_base.ckpt

Though I was getting weird results where it looked like it processed cropped rectangular blocks of the image one at a time, causing abrupt transitions in the depth estimate between the blocks. Probably has something to do with the sliding window thing / batch size / some cropping thing, but I haven't figured that out yet.