Evaluation of detection proposal algorithms. The code belongs to the BMVC paper How good are detection proposals, really? and an upcoming journal paper. Have a look at the Project Page for more information.
Please contact me if you're having trouble with the code!
- Get the data that you want to use from the Project Page.
- Edit
get_config.m
to point to the right locations for images, candidates and so on. - Make sure you either start matlab in the root directory of the code or run
startup.m
manually once. - Run
plot_recall_voc07.m
, curves will be in the figures subdirectory.
- Follow Plot evaluation curves.
- Write a wrapper function that takes an image and the number of proposal boxes and returns the proposals and scores. Proposals are a nx4 matrix, where n is the number of proposals and every row has the format
[x1 x2 y1 y2]
(x and y are 1-based image coordinates). Seemethod_wrappers/
for examples. - Add your method to
shared/get_method_configs.m
- Run
compute_recall_candidates_voc07.m
passing only the config of your method as an argument. If your method is slow, you probably want to parallelize it in a cluster. - Run
plot_recall_voc07.m
, curves will be in the figures subdirectory.
You don't have to use compute_recall_candidates_voc07.m
, but you can have a look to get an idea about how to save the candidates in the right format so, plot_recall_voc07.m
will be able to read it.