unsky
unsky
@makefile， Your understanding is very shallow. you should read the codes carefully. i dont use resize,in the code, resize will have lots of problem. i use pad with image stride....
FPN里面并没有太考虑模型的参数量，有大量的可优化空间，但不属于该论文的讨论范畴
https://github.com/msracver/Deformable-ConvNets mxnet version. https://github.com/unsky/FPN caffe version.
@yian2271368
i explore the method in proposal + classification, not the regression(like ssd), the initialization is a method for early training to balanced the heavy inbalanced examples, but my goal is...
@Franciszzj i think, there are some bug in your code, you can check the prob cls value and gradient value in the code,,, and i was exploring the performance in...
@zimenglan-sysu-512 --- mxnet (faster rcnn) parameters: MXNET_VERSION: "mxnet" output_path: "./output/rcnn/kitti" symbol: resnet_v1_101_rcnn_dcn gpus: '1,2,3,4,5,6,7' CLASS_AGNOSTIC: false SCALES: - 740 - 2448 default: frequent: 100 kvstore: device network: pretrained: "./model/pretrained_model/resnet_v1_101" pretrained_epoch:...
@zimenglan-sysu-512 e。。http://www.cvlibs.net/datasets/kitti/eval_object.php?obj_benchmark=2d
@iFighting focal loss is a method to process the imbalance examples. it is not better in the balance scene. for the level of imbalance , you must choose a suit...
i do not understand your mean, the codes in metric? the loss metric just a display and dont affact the model. if your loss_ is in the metric.py pro_ and...