search

s-gupta / rcnn-depth

Learning Rich Features from RGB-D Images for Object Detection and Segmentation
BSD 2-Clause "Simplified" License
365 stars 150 forks source link

Evaluation script #3

Closed varun-nagaraja closed 9 years ago

varun-nagaraja commented 9 years ago

I have been able to obtain object detections and now I would like to evaluate the detections as done in the paper. Could you tell me how I can obtain the APb and APr scores?

s-gupta commented 9 years ago

Hi @varun-nagaraja

If you did the following to train the model: res = rcnn_all('task-detection', 'rgb_hha', 'trainval', 'test');, then the variable res should contain the APb. It is also stored in the detector directory (in fullfile(p.cache_dir, 'detection', 'detector', 'rgb_hha_30000_trainval', 'pr-curves', 'results_nyud2_test_release.mat').

We haven't yet released code for instance segmentation. You will need that to compute the instance segmentation output (and to get APr). If you just want the benchmarking code to compute APr and use it to benchmark your algorithm, I can try to clean that part first and release that, let me know.

varun-nagaraja commented 9 years ago

Got it, thanks! I will just need APb for now.