Jian Ding

On the construct of my OR-Resnet, I can get a normal accuracy if I do not do random rotation operation on the test set. But when I random rotate the...

@swz30 @MaxBazik @swz30 How many gpus you used? The learning rate need to be set according to the number of gpus. Our default setting is for 4 gpus. See this...

Since the train.txt and test.txt represent the names of split images, different split images would get different train.txt and test.txt. I will update the DOTA_devkit later. You can write a...

I encountered the issue in another Faster-RCNN repo(https://github.com/jwyang/faster-rcnn.pytorch). The analytical gradients of the output from the gradcheck function in pytorch are all zero. But when I print the gradient results...

> #137 sovled this problem. Hi, I tried the version after it. But it seems the PR did not fix the problem. I used two Titan X Pascal to train....

> @dingjiansw101 A better practice is here: https://gluon-cv.mxnet.io Is FPN and deformable available in this practice?

Thanks for the reply. But I still have a little question about (1). From my understanding, the Transformer Layers before the Grouping Block already have the information propagations. Why do...

The learning rate of SGD depends on the batch size. If your batch size is bs, then the lr = 0.01/8 * bs.

You should submit the zip of Task1_results_nms or Transed_Task2_results_nms to the evaluation server (https://captain-whu.github.io/DOTA/evaluation.html).

Please see https://github.com/dingjiansw101/AerialDetection/blob/master/GETTING_STARTED.md for reference.