qimw
qimw
Hi all, I am trying to add lighthead module on faster rcnn. I add a new light_branch on faster_rcnn_heads.py which conbines the functions of box_head and box_out. This module needs...
I am adapting from light dataset to a darker one. When I am running test on the source domain, I found that the performance is highly degraded if performing model.eval()....
您好， 感谢您的开源工作！我在尝试复现多尺度训练测试的 R-50-FPN 结果，但是只达到了79.39%，麻烦您帮忙看一下是否有哪里出错： 1.根据DOTA1.0 ms_trainval.json分割训练验证集，根据ms_test.json分割测试集，分别是68325张和71888张图 2.使用faster_rcnn_orpn_r50_fpn_1x_ms_rr_dota10单卡训练 3.使用tools/test.py 得到提交压缩包 最终服务器返回测试结果为： mAP: 0.7939351339104009 ap of each class: plane:0.8942105218685668, baseball-diamond:0.8374610362955522, bridge:0.5827316700115821, ground-track-field:0.7559930104103494, small-vehicle:0.7582806851195398, large-vehicle:0.8308442404507247, ship:0.8861535597433863, tennis-court:0.9088850209713762, basketball-court:0.8643310439273802, storage-tank:0.87514554997598, soccer-ball-field:0.7221988431931634, roundabout:0.7194164215729385, harbor:0.8121602830954224, swimming-pool:0.7630957318293694,...
In the config.py, the TRAIN.nr_ohem_sampling = 256 * train_batch_per_gpu. The TRAIN.RPN_BATCHSIZE is also 256. Did you really use ohem in your code ?
The current code implement the Slice operation through F.grid_sample(). But I found the indexes may be wrong. ``` import torch import torch.nn.functional as F import numpy as np box =...