add windows 10 support

Author: traveller59

README.md

@@ -1,7 +1,7 @@
 # SECOND-V1.5 for KITTI object detection
 SECOND-V1.5 detector.
 
-ONLY support python 3.6+, pytorch 1.0.0+. Tested in Ubuntu 16.04/18.04.
+ONLY support python 3.6+, pytorch 1.0.0+. Tested in Ubuntu 16.04/18.04/Windows 10.
 
 ## News
 
@@ -53,13 +53,17 @@ cd ./second.pytorch/second
 It is recommend to use Anaconda package manager.
 
 ```bash
-pip install shapely fire pybind11 tensorboardX protobuf scikit-image numba pillow
+conda install scikit-image scipy numba pillow matplotlib
+```
+
+```bash
+pip install fire tensorboardX protobuf opencv-python
 ```
 
 If you don't have Anaconda:
 
 ```bash
-pip install numba
+pip install numba scikit-image scipy pillow
 ```
 
 Follow instructions in [spconv](https://github.com/traveller59/spconv) to install spconv. 

RELEASE.md

@@ -18,4 +18,5 @@ points([N, 4])->voxels([N, 5, 4])->Features([N, 4])->Sparse Convolution Networks
 * Eval functions now support custom data evaluation. you need to specify z_center and z_axis in eval function.
 2. Better RPN, you can add custom block by inherit RPNBase and implement _make_layer method.
 3. Update pretrained model.
-4. Add a simple inference notebook. everyone should start this project by that notebook.
+4. Add a simple inference notebook. everyone should start this project by that notebook.
+5. Add windows support. Training on windows is slow than linux.

second/core/preprocess.py

@@ -6,7 +6,6 @@
 
 import numba
 import numpy as np
-from shapely.geometry import Polygon
 
 from second.core import box_np_ops
 from second.core.geometry import (points_in_convex_polygon_3d_jit,

second/kittiviewer/backend.py

@@ -21,7 +21,6 @@
 import matplotlib.pyplot as plt
 import numba
 import skimage
-from shapely.geometry import Polygon
 from skimage import io
 
 import second.core.box_np_ops as box_np_ops

second/pytorch/models/rpn.py

@@ -2,7 +2,6 @@
 from enum import Enum
 
 import numpy as np
-import sparseconvnet as scn
 import torch
 from torch import nn
 from torch.nn import functional as F

second/pytorch/models/voxelnet.py

@@ -3,7 +3,6 @@
 from functools import reduce
 
 import numpy as np
-import sparseconvnet as scn
 import torch
 from torch import nn
 from torch.nn import functional as F

second/pytorch/train.py

@@ -78,6 +78,11 @@ def build_network(model_cfg, measure_time=False):
     net = second_builder.build(model_cfg, voxel_generator, target_assigner, measure_time=measure_time)
     return net
 
+def _worker_init_fn(worker_id):
+    time_seed = np.array(time.time(), dtype=np.int32)
+    np.random.seed(time_seed + worker_id)
+    print(f"WORKER {worker_id} seed:", np.random.get_state()[1][0])
+
 def train(config_path,
           model_dir,
           result_path=None,
@@ -171,10 +176,6 @@ def train(config_path,
         training=False,
         voxel_generator=voxel_generator,
         target_assigner=target_assigner)
-    def _worker_init_fn(worker_id):
-        time_seed = np.array(time.time(), dtype=np.int32)
-        np.random.seed(time_seed + worker_id)
-        print(f"WORKER {worker_id} seed:", np.random.get_state()[1][0])
 
     dataloader = torch.utils.data.DataLoader(
         dataset,