Release

config.py

@@ -1,6 +1,7 @@
 """Configurate arguments."""
 import argparse
 
+# TODO: reproduce all number
 
 INPUT_IMAGE_SIZE = 512
 # 0: confidence, 1: point_shape, 2: offset_x, 3: offset_y, 4: cos(direction),
@@ -20,7 +21,10 @@ VSLOT_MAX_DIST = 0.1099427457599304
 HSLOT_MIN_DIST = 0.15057789144568634
 HSLOT_MAX_DIST = 0.44449496544202816
 
-# angle_prediction_error = 0.1384059287593468
+SHORT_SEPARATOR_LENGTH = 0.199519231
+LONG_SEPARATOR_LENGTH = 0.46875
+
+# angle_prediction_error = 0.1384059287593468 collected from evaluate.py
 BRIDGE_ANGLE_DIFF = 0.09757113548987695 + 0.1384059287593468
 SEPARATOR_ANGLE_DIFF = 0.284967562063968 + 0.1384059287593468
 
@@ -52,7 +56,7 @@ def get_parser_for_training():
                         help="Batch size.")
     parser.add_argument('--data_loading_workers', type=int, default=32,
                         help="Number of workers for data loading.")
-    parser.add_argument('--num_epochs', type=int, default=10,
+    parser.add_argument('--num_epochs', type=int, default=20,
                         help="Number of epochs to train for.")
     parser.add_argument('--lr', type=float, default=1e-4,
                         help="The learning rate of back propagation.")
@@ -67,10 +71,6 @@ def get_parser_for_evaluation():
     parser = argparse.ArgumentParser()
     parser.add_argument('--dataset_directory', required=True,
                         help="The location of dataset.")
-    parser.add_argument('--batch_size', type=int, default=32,
-                        help="Batch size.")
-    parser.add_argument('--data_loading_workers', type=int, default=64,
-                        help="Number of workers for data loading.")
     parser.add_argument('--enable_visdom', action='store_true',
                         help="Enable Visdom to visualize training progress")
     add_common_arguments(parser)
@@ -89,6 +89,7 @@ def get_parser_for_ps_evaluation():
     add_common_arguments(parser)
     return parser
 
+
 def get_parser_for_inference():
     """Return argument parser for inference."""
     parser = argparse.ArgumentParser()

data/__init__.py

@@ -1,4 +1,4 @@
 """Data related package."""
-from .data_process import get_predicted_points, pair_marking_points, pass_through_third_point
+from .process import get_predicted_points, pair_marking_points, pass_through_third_point
 from .dataset import ParkingSlotDataset
 from .struct import MarkingPoint, Slot, match_marking_points, match_slots, calc_point_squre_dist, calc_point_direction_angle

data/process.py

@@ -1,4 +1,4 @@
-"""Defines data structure and related function to process these data."""
+"""Defines related function to process defined data structure."""
 import math
 import numpy as np
 import torch

evaluate.py

@@ -1,11 +1,9 @@
 """Evaluate directional marking point detector."""
-import cv2 as cv
 import torch
 import config
 import util
 from data import get_predicted_points, match_marking_points, calc_point_squre_dist, calc_point_direction_angle
 from data import ParkingSlotDataset
-from inference import plot_points
 from model import DirectionalPointDetector
 from train import generate_objective
 
@@ -74,12 +72,6 @@ def evaluate_detector(args):
         pred_points = get_predicted_points(prediction[0], 0.01)
         predictions_list.append(pred_points)
 
-        # if not is_gt_and_pred_matched(marking_points, pred_points,
-        #                               config.CONFID_THRESH_FOR_POINT):
-        #     cvimage = util.tensor2array(image[0]).copy()
-        #     plot_points(cvimage, pred_points)
-        #     cv.imwrite('flaw/%d.jpg' % iter_idx, cvimage,
-        #                [int(cv.IMWRITE_JPEG_QUALITY), 100])
         dists, angles = collect_error(marking_points, pred_points,
                                       config.CONFID_THRESH_FOR_POINT)
         position_errors += dists

inference.py

@@ -60,10 +60,15 @@ def plot_slots(image, pred_points, slots):
         p1_y = height * point_b.y - 0.5
         vec = np.array([p1_x - p0_x, p1_y - p0_y])
         vec = vec / np.linalg.norm(vec)
-        p2_x = p0_x + 200*vec[1]
+        distance = calc_point_squre_dist(point_a, point_b)
-        p2_y = p0_y - 200*vec[0]
+        if config.VSLOT_MIN_DIST <= distance <= config.VSLOT_MAX_DIST:
-        p3_x = p1_x + 200*vec[1]
+            separating_length = config.LONG_SEPARATOR_LENGTH
-        p3_y = p1_y - 200*vec[0]
+        elif config.HSLOT_MIN_DIST <= distance <= config.HSLOT_MAX_DIST:
+            separating_length = config.SHORT_SEPARATOR_LENGTH
+        p2_x = p0_x + height * separating_length * vec[1]
+        p2_y = p0_y - width * separating_length * vec[0]
+        p3_x = p1_x + height * separating_length * vec[1]
+        p3_y = p1_y - width * separating_length * vec[0]
         p0_x = int(round(p0_x))
         p0_y = int(round(p0_y))
         p1_x = int(round(p1_x))
@@ -122,7 +127,7 @@ def detect_video(detector, device, args):
     frame_height = int(input_video.get(cv.CAP_PROP_FRAME_HEIGHT))
     output_video = cv.VideoWriter()
     if args.save:
-        output_video.open('record.avi', cv.VideoWriter_fourcc(*'MJPG'),
+        output_video.open('record.avi', cv.VideoWriter_fourcc(*'XVID'),
                           input_video.get(cv.CAP_PROP_FPS),
                           (frame_width, frame_height), True)
     frame = np.empty([frame_height, frame_width, 3], dtype=np.uint8)

model/detector.py

@@ -55,8 +55,7 @@ class DirectionalPointDetector(nn.modules.Module):
         self.predict = nn.Sequential(*layers)
 
     def forward(self, *x):
-        feature = self.extract_feature(x[0])
+        prediction = self.predict(self.extract_feature(x[0]))
-        prediction = self.predict(feature)
         # 4 represents that there are 4 value: confidence, shape, offset_x,
         # offset_y, whose range is between [0, 1].
         point_pred, angle_pred = torch.split(prediction, 4, dim=1)

ps_evaluate.py

@@ -12,6 +12,7 @@ from inference import detect_marking_points, inference_slots
 
 
 def get_ground_truths(label):
+    """Read label to get ground truth slot."""
     slots = np.array(label['slots'])
     if slots.size == 0:
         return []

train.py

@@ -33,8 +33,8 @@ def generate_objective(marking_points_batch, device):
     gradient[:, 0].fill_(1.)
     for batch_idx, marking_points in enumerate(marking_points_batch):
         for marking_point in marking_points:
-            col = math.floor(marking_point.x * 16)
+            col = math.floor(marking_point.x * config.FEATURE_MAP_SIZE)
-            row = math.floor(marking_point.y * 16)
+            row = math.floor(marking_point.y * config.FEATURE_MAP_SIZE)
             # Confidence Regression
             objective[batch_idx, 0, row, col] = 1.
             # Makring Point Shape Regression
@@ -76,11 +76,11 @@ def train_detector(args):
                              collate_fn=lambda x: list(zip(*x)))
 
     for epoch_idx in range(args.num_epochs):
-        for iter_idx, (image, marking_points) in enumerate(data_loader):
+        for iter_idx, (images, marking_points) in enumerate(data_loader):
-            image = torch.stack(image).to(device)
+            images = torch.stack(images).to(device)
 
             optimizer.zero_grad()
-            prediction = dp_detector(image)
+            prediction = dp_detector(images)
             objective, gradient = generate_objective(marking_points, device)
             loss = (prediction - objective) ** 2
             loss.backward(gradient)
@@ -89,7 +89,7 @@ def train_detector(args):
             train_loss = torch.sum(loss*gradient).item() / loss.size(0)
             logger.log(epoch=epoch_idx, iter=iter_idx, train_loss=train_loss)
             if args.enable_visdom:
-                plot_prediction(logger, image, marking_points, prediction)
+                plot_prediction(logger, images, marking_points, prediction)
         torch.save(dp_detector.state_dict(),
                    'weights/dp_detector_%d.pth' % epoch_idx)
         torch.save(optimizer.state_dict(), 'weights/optimizer.pth')

util/log.py

@@ -1,4 +1,4 @@
-# -*- coding: utf-8 -*-
+"""Class for logging."""
 import math
 import numpy as np
 from visdom import Visdom

util/utils.py

@@ -30,6 +30,8 @@ class Timer(object):
 
     def calc_average_time(self):
         """Calculate average elapsed time of timer."""
+        if self.count == 0:
+            return 0.
         return self.total_time / self.count