Ace
In order to avoid uncontiguous problem（most operation in pytorch is contiguous）. You can find there is a [contiguous check](https://github.com/pytorch/extension-cpp/blob/master/cuda/lltm_cuda.cpp) in the c++ code, so in order to "remove" the uncontiguous...
1. the batch_index means: which batch does this roi belong to. (for example, if there only one "object", the batch_index is always 0) 2. you can reference [simple-faster-rcnn-pytorch](https://github.com/ruotianluo/pytorch-faster-rcnn/blob/master/lib/roi_data_layer/layer.py), [faster-rcnn.pytorch](https://github.com/jwyang/faster-rcnn.pytorch/tree/master/lib/roi_data_layer) to...
In pytorch 0.4.1, you need to install from source and include the head file and .so file in cmakelist. However, there are an easy way in 1.0 (now preview 1.0),...
Would you mind explain the value you use in random_shape? (In your code, this is `(np.random.rand() - 0.5) * np.random.exponential(0.4)`)，there are some thing confuse me: 1. this value will cause...
After read the shepp logan code. ("force" modify the value in `_modified_shepp_logan_ellipsoids`). I am curious about the guideline of the code `random_phantom`.
The ECSSD dataset have already divided the train, val and test. if you want to train on ECSSD please using follow lines: https://github.com/AceCoooool/NLDF-pytorch/blob/7c4a5dddab277c6136378e1592ef7c287b60d314/main.py#L40
BaiDuYun：链接: https://pan.baidu.com/s/1_K-uCyS3yETzTxfnZNz3Aw 提取码: w4gq I am not sure whether it's divided by myself ? (I do not update this resposity for a long time, I am sorry)
1. You can download the weights and cfg from [offical website](https://pjreddie.com/darknet/yolo/) 2. Due to the line of `sys.path.append('..')`: You should run it under the directory of tools (You can delete...
The model is built as the paper described. (Only training stragety modified in this implementation) So I think the speed is nearly the same as the paper. (I will test...
I will update the pre-trained model (the older version is pytorch0.3, and the bn layers is changed after 0.4.1) You can download with this link: https://drive.google.com/open?id=1v2LRWjfmbQc8VREX39LopX4v-Rd3YEOm > this performance may...