modify .gitignore and readme

.gitignore

@@ -99,3 +99,9 @@ ENV/
 
 # mypy
 .mypy_cache/
+
+# apple file
+.DS_Store
+
+# vscode
+.vscode

README.md

@@ -6,12 +6,10 @@ This is a repository for a [Deformable Convolution](https://arxiv.org/abs/1703.0
 
 Tensorflow(with GPU configured)
 
-cuda 8.0
+Cuda 8.0
 
 g++ 4.9.2
 
-
-
 *Note*: Only tested on platform where corresponding version of g++ and cuda installed, other version might generally be fine, but may need to modify the compile script.
 
 ## Usage
@@ -21,8 +19,6 @@ g++ 4.9.2
 3. `import lib.deform_conv_op as deform_conv_op` in your python script (make sure PYTHON_PATH was set currectly).
 
 
-
-
 ## TODO
 
 - [x] Basic test with original implementation.
@@ -32,4 +28,8 @@ g++ 4.9.2
 
 - [ ] Some demo and visualization.
 - [ ] Backward time costs too much.
-- [ ] Other ops.
+- [ ] Other ops.
+
+## Benchmark
+
+Benchmark script was borrowed from [here](https://github.com/soumith/convnet-benchmarks/blob/master/tensorflow/benchmark_alexnet.py). The forward time is fine, for 100x3x224x224 data, it runs about in 0.077s. But backward time generaly undesired, it cost 0.558s to run a batch of same data. Note I write all backward of three inputs(data, offset, kernels) together, rather than like many tensorflow conv ops spliting input_backwards and kernel_backwards to two ops, so this might be one of the reason. In addition, because  sometimes I find it hard to manipulate `tensorflow::Tensor` , I write a simple cuda kernel that does nothing but add one tensor to another, for accumulating gradients along batch in kernel gradient implementation, don't know whether it affects performance.