Hi, @ChenhongyiYang
Thanks for providing the detailed code of your work.
I tried running your code for COCO and was not able to reproduce the results. Following is the result:
Round 1: OrderedDict([('bbox_mAP', 0.062), ('bbox_mAP_50', 0.116), ('bbox_mAP_75', 0.057), ('bbox_mAP_s', 0.025), ('bbox_mAP_m', 0.062), ('bbox_mAP_l', 0.081), ('bbox_mAP_copypaste', '0.062 0.116 0.057 0.025 0.062 0.081')])
Round 2: OrderedDict([('bbox_mAP', 0.124), ('bbox_mAP_50', 0.22), ('bbox_mAP_75', 0.123), ('bbox_mAP_s', 0.055), ('bbox_mAP_m', 0.133), ('bbox_mAP_l', 0.161), ('bbox_mAP_copypaste', '0.124 0.220 0.123 0.055 0.133 0.161')])
Round 3: OrderedDict([('bbox_mAP', 0.149), ('bbox_mAP_50', 0.256), ('bbox_mAP_75', 0.151), ('bbox_mAP_s', 0.067), ('bbox_mAP_m', 0.163), ('bbox_mAP_l', 0.193), ('bbox_mAP_copypaste', '0.149 0.256 0.151 0.067 0.163 0.193')])
Round 4: OrderedDict([('bbox_mAP', 0.174), ('bbox_mAP_50', 0.295), ('bbox_mAP_75', 0.178), ('bbox_mAP_s', 0.078), ('bbox_mAP_m', 0.187), ('bbox_mAP_l', 0.228), ('bbox_mAP_copypaste', '0.174 0.295 0.178 0.078 0.187 0.228')])
I think it might be the issue with the initial random selection. Whats your take on it,
Also is it possible for you to share your best selection json files? Coz I tried two cycle for AL and the results were kind of similar in this range.
Hi, @ChenhongyiYang Thanks for providing the detailed code of your work. I tried running your code for COCO and was not able to reproduce the results. Following is the result: Round 1: OrderedDict([('bbox_mAP', 0.062), ('bbox_mAP_50', 0.116), ('bbox_mAP_75', 0.057), ('bbox_mAP_s', 0.025), ('bbox_mAP_m', 0.062), ('bbox_mAP_l', 0.081), ('bbox_mAP_copypaste', '0.062 0.116 0.057 0.025 0.062 0.081')]) Round 2: OrderedDict([('bbox_mAP', 0.124), ('bbox_mAP_50', 0.22), ('bbox_mAP_75', 0.123), ('bbox_mAP_s', 0.055), ('bbox_mAP_m', 0.133), ('bbox_mAP_l', 0.161), ('bbox_mAP_copypaste', '0.124 0.220 0.123 0.055 0.133 0.161')]) Round 3: OrderedDict([('bbox_mAP', 0.149), ('bbox_mAP_50', 0.256), ('bbox_mAP_75', 0.151), ('bbox_mAP_s', 0.067), ('bbox_mAP_m', 0.163), ('bbox_mAP_l', 0.193), ('bbox_mAP_copypaste', '0.149 0.256 0.151 0.067 0.163 0.193')]) Round 4: OrderedDict([('bbox_mAP', 0.174), ('bbox_mAP_50', 0.295), ('bbox_mAP_75', 0.178), ('bbox_mAP_s', 0.078), ('bbox_mAP_m', 0.187), ('bbox_mAP_l', 0.228), ('bbox_mAP_copypaste', '0.174 0.295 0.178 0.078 0.187 0.228')])
I think it might be the issue with the initial random selection. Whats your take on it, Also is it possible for you to share your best selection json files? Coz I tried two cycle for AL and the results were kind of similar in this range.