Edgeboxes Proposal and DPM connections

After several days of reading the literature and code, I have a certain understanding of edgeboxes and DPM.

And now it's decided to simply connect them.

That is Edgeboxes proposal recommended window and then DPM to judge

But as I understand DPM, the window recommendations do not work very well. Because DPM needs a sliding match inside.

If you use the Superpixel recommendation to directly overwrite the DPM effect internal matching process, the effect should be more obvious.

After an afternoon of blind writing, tinkering with

Finally connected, and found that the performance has become worse.

It would have been 3 seconds. =_=

Hesitation Edgeboxes recommended the first few windows are generally larger, and overlap serious. But the target will be inside, so the first few windows, DPM will detect the target.

But if there is no target, it is false, then hehe.

The DMP for each window is calculated once hog, and then the match is slid.

and went to see what makes for effective detection proposals?

Found inside in order to test various recommended methods, first use the classic sliding window, and then use a variety of recommended methods.

Also said a sentence this does not speed-up detection, but enables evaluating the effect of proposals on detection quality.

It's really for testing.

Edgeboxes Proposal and DPM connections