Some tweaks

README.md

@@ -12,16 +12,21 @@ This is an unofficial inplementation of [VoxelNet: End-to-End Learning for Point
 
 # Installation
 1. Clone this repository.
-2. Compile the Cython module
+2. Install the necessary packages 
+```bash
+$ sudo apt-get install libboost-all-dev
+$ sudo apt-get install python-numpy
+```
+3. Compile the Cython module
 ```bash
 $ python3 setup.py build_ext --inplace
 ```
-3. Compile the evaluation code
+4. Compile the evaluation code
 ```bash
 $ cd kitti_eval
 $ g++ -o evaluate_object_3d_offline evaluate_object_3d_offline.cpp
 ```
-4. grant the execution permission to evaluation script
+5. grant the execution permission to evaluation script
 ```bash
 $ cd kitti_eval
 $ chmod +x launch_test.sh
@@ -66,9 +71,6 @@ Training on two Nvidia 1080 Ti GPUs takes around 3 days (160 epochs as reported
 $ python3 parse_log.py predictions
 ```
 
-4. There is a pre-trained model for car in `save_model/pre_trained_car`.
-
-
 # Evaluate
 1. run `test.py -n default` to produce final predictions on the validation set after training is done. Change `-n` flag to `pre_trained_car` will start testing for the pre-trained model (only car model provided for now).
 ```bash

config.py

@@ -18,12 +18,12 @@
 cfg = __C
 
 # for dataset dir
-__C.DATA_DIR = '/media/hdc/KITTI/for_voxelnet/cropped_dataset'
-__C.CALIB_DIR = '/media/hdc/KITTI/calib/data_object_calib/training/calib'
+__C.DATA_DIR = '/home/anshul/Project/voxelnet-1/data_dir'
+__C.CALIB_DIR = '/home/anshul/Project/voxelnet-1/data_object_calib/training/calib'
 
 
 # for gpu allocation
-__C.GPU_AVAILABLE = '0,1'
+__C.GPU_AVAILABLE = '0,1,2'
 __C.GPU_USE_COUNT = len(__C.GPU_AVAILABLE.split(','))
 __C.GPU_MEMORY_FRACTION = 1
 

data/crop.py

@@ -1,6 +1,7 @@
 import numpy as np
-from scipy.misc import imread
-
+import PIL
+from imageio import imread
+#from scipy.misc.pilutil import imread
 CAM = 2
 
 def load_velodyne_points(filename):
@@ -50,7 +51,7 @@ def project_velo_points_in_img(pts3d, T_cam_velo, Rrect, Prect):
 
 
 def align_img_and_pc(img_dir, pc_dir, calib_dir):
-    
+
     img = imread(img_dir)
     pts = load_velodyne_points( pc_dir )
     P, Tr_velo_to_cam, R_cam_to_rect = load_calib(calib_dir)
@@ -79,9 +80,9 @@ def align_img_and_pc(img_dir, pc_dir, calib_dir):
     return points
 
 # update the following directories
-IMG_ROOT = '/media/hdc/KITTI/image/training/image_2/'
-PC_ROOT = '/media/hdc/KITTI/point_cloud/raw_bin_files/training/velodyne/'
-CALIB_ROOT = '/media/hdc/KITTI/calib/data_object_calib/training/calib/'
+IMG_ROOT = "/home/anshul/Project/voxelnet-1/data_object_image_2/training/image_2/"
+PC_ROOT = "/home/anshul/Project/voxelnet-1/data_object_velodyne/training/velodyne/"
+CALIB_ROOT = "/home/anshul/Project/voxelnet-1/data_object_calib/training/calib/"
 
 
 
@@ -91,12 +92,8 @@ def align_img_and_pc(img_dir, pc_dir, calib_dir):
     calib_dir = CALIB_ROOT + '%06d.txt' % frame
 
     points = align_img_and_pc(img_dir, pc_dir, calib_dir)
-
-    output_name = PC_ROOT + frame + '.bin'
-    points[:,:4].astype('float32').tofile(output_name)
-
-
-
-
-
 
+    #output_name = PC_ROOT + str(frame) + '.bin'
+    output_name = PC_ROOT + str('%06d' % frame) + '.bin'
+    print(frame)
+    points[:,:4].astype('float32').tofile(output_name)

datasplit.py

@@ -0,0 +1,28 @@
+import os
+import shutil
+
+#dir="/home/anshul/Project/voxelnet-1/data_object_velodyne/training/velodyne"
+#dir="/home/anshul/Project/voxelnet-1/data_object_label_2/training/label_2"
+dir="/home/anshul/Project/voxelnet-1/data_object_image_2/training/image_2"
+val_find="/home/anshul/Project/voxelnet-1/val.txt"
+#dest="/home/anshul/Project/voxelnet-1/data_object_velodyne/validation/velodyne"
+#dest="/home/anshul/Project/voxelnet-1/data_object_label_2/validation/label_2"
+dest="/home/anshul/Project/voxelnet-1/data_object_image_2/validation/image_2"
+
+#dest="D:\VoxelNet-tensorflow\data_object_velodyne\\train\\velodyne"
+#dest="D:\VoxelNet-tensorflow\data_object_label_2\\train\label_2"
+#dest="D:\VoxelNet-tensorflow\data_object_image_2\\train\image_2"
+
+num=[]
+with open(val_find) as vf:
+    for i in vf:
+        i=i.split()
+        #num.append(i[0]+'.bin')
+        #num.append(i[0]+'.txt')
+        num.append(i[0]+'.png')
+print(dest)
+
+os.makedirs(dest)
+for i in num:
+    shutil.move(os.path.join(dir,i),dest)
+#shutil.move('D:\\000000.bin',dest)

kitti_eval/launch_test.sh

@@ -1,5 +1,5 @@
 # crooped gt dir
-GT_DIR=/media/hdc/KITTI/for_voxelnet/cropped_dataset/validation/label_2
+GT_DIR=/home/anshul/Project/voxelnet-1/data_dir/validation/label_2
 # pred dir
 PRED_DIR=$1
 

model/group_pointcloud.py

@@ -14,18 +14,18 @@ class VFELayer(object):
     def __init__(self, out_channels, name):
         super(VFELayer, self).__init__()
         self.units = int(out_channels / 2)
-        with tf.variable_scope(name, reuse=tf.AUTO_REUSE) as scope:
-            self.dense = tf.layers.Dense(
-                self.units, tf.nn.relu, name='dense', _reuse=tf.AUTO_REUSE, _scope=scope)
-            self.batch_norm = tf.layers.BatchNormalization(
-                name='batch_norm', fused=True, _reuse=tf.AUTO_REUSE, _scope=scope)
+        with tf.compat.v1.variable_scope(name, reuse=tf.compat.v1.AUTO_REUSE) as scope:
+            self.dense = tf.compat.v1.layers.Dense(
+                self.units, tf.compat.v1.nn.relu, name='dense', _reuse=tf.compat.v1.AUTO_REUSE, _scope=scope)
+            self.batch_norm = tf.compat.v1.layers.BatchNormalization(
+                name='batch_norm', fused=True, _reuse=tf.compat.v1.AUTO_REUSE, _scope=scope)
 
     def apply(self, inputs, mask, training):
         # [K, T, 7] tensordot [7, units] = [K, T, units]
         pointwise = self.batch_norm.apply(self.dense.apply(inputs), training)
 
         #n [K, 1, units]
-        aggregated = tf.reduce_max(pointwise, axis=1, keep_dims=True)
+        aggregated = tf.reduce_max(pointwise, axis=1, keepdims=True)
 
         # [K, T, units]
         repeated = tf.tile(aggregated, [1, cfg.VOXEL_POINT_COUNT, 1])
@@ -49,30 +49,28 @@ def __init__(self, training, batch_size, name=''):
         # scalar
         self.batch_size = batch_size
         # [ΣK, 35/45, 7]
-        self.feature = tf.placeholder(
+        self.feature = tf.compat.v1.placeholder(
             tf.float32, [None, cfg.VOXEL_POINT_COUNT, 7], name='feature')
         # [ΣK]
-        self.number = tf.placeholder(tf.int64, [None], name='number')
+        self.number = tf.compat.v1.placeholder(tf.int64, [None], name='number')
         # [ΣK, 4], each row stores (batch, d, h, w)
-        self.coordinate = tf.placeholder(
+        self.coordinate = tf.compat.v1.placeholder(
             tf.int64, [None, 4], name='coordinate')
 
-        with tf.variable_scope(name, reuse=tf.AUTO_REUSE) as scope:
+        with tf.compat.v1.variable_scope(name, reuse=tf.compat.v1.AUTO_REUSE) as scope:
             self.vfe1 = VFELayer(32, 'VFE-1')
             self.vfe2 = VFELayer(128, 'VFE-2')
 
         # boolean mask [K, T, 2 * units]
-        mask = tf.not_equal(tf.reduce_max(
-            self.feature, axis=2, keep_dims=True), 0)
+        mask = tf.compat.v1.not_equal(tf.reduce_max(
+            self.feature, axis=2, keepdims=True), 0)
         x = self.vfe1.apply(self.feature, mask, self.training)
         x = self.vfe2.apply(x, mask, self.training)
 
         # [ΣK, 128]
-        voxelwise = tf.reduce_max(x, axis=1)
+        voxelwise = tf.compat.v1.reduce_max(x, axis=1)
 
         # car: [N * 10 * 400 * 352 * 128]
         # pedestrian/cyclist: [N * 10 * 200 * 240 * 128]
-        self.outputs = tf.scatter_nd(
+        self.outputs = tf.compat.v1.scatter_nd(
             self.coordinate, voxelwise, [self.batch_size, 10, cfg.INPUT_HEIGHT, cfg.INPUT_WIDTH, 128])
-
-