pva-faster-rcnn icon indicating copy to clipboard operation
pva-faster-rcnn copied to clipboard

Published 20 hours ago • verified publisher icon sanghoon

MSCOCO mAP

Open siddharthm83 opened this issue 8 years ago • 5 comments

Hi, this is not a code issue but I was curious if you have mAP numbers on COCO. The paper only has pascal VOC results.

siddharthm83 avatar Dec 22 '16 21:12 siddharthm83

Here are my intermediate results. I will update it again when I complete 500k training iteration. results

hengck23 avatar Dec 27 '16 06:12 hengck23

Since our target was to develop a detector for few selected objects, we haven't evaluated our models on the MS COCO (80 class) dataset.

sanghoon avatar Dec 27 '16 06:12 sanghoon

Here is the updated and final results for your reference. Note that I do not try the best parameters here and if we do some fine adjustment, the results could get better. results_140

hengck23 avatar Dec 28 '16 12:12 hengck23

more updated here! Note that the train data i used for pvanet-full is less than that of other methods in comparsion. Maybe I need to redo the experiments later. (there is typo error: plateau windows should be 50k,50k,50k)

pic5

pic6

pic1

pic2

hengck23 avatar Jan 01 '17 12:01 hengck23

how about the resolution of the image's short edge for coco test data?

hectorgui avatar Sep 26 '17 02:09 hectorgui