Analysis of Parking Spaces

Structure

.
├── network/
|   ├── alexnet.py
|   ├── lenet.py
|   ├── vgg7.py
|   ├── vgg13.py
|   ├── vgg13v.py
|   └── vgg16.py
├── src_img/
|   ├── PUCPR/
|   ├── UFPR04/
|   └── UFPR05/
├── train_data/
|   ├── models/
|   ├── test_img/
|   ├── test_seg/
|   └── train/
├── calc_accuracy.py
├── data_generator.py
├── detect_contour.py
├── test_image.py
├── test_segment.py
├── train_network.py
└── README.md

Hardware requirements

NVIDIA® GPU card with CUDA® Compute Capability 3.5 or higher.

Software requirements

• GPU drivers (410.x)
• CUDA (9.0)
• cuDNN (≥7.4.1 for CUDA 9.0)
• Anaconda

Datasets

• Parking Lot Database
  • PUCPR: 4474 pics
  • UFPR04: 3791 pics
  • UFPR05: 4152 pics

Procedure

0. Debug

Related files:

• test_segment.py
• train_data/models/
• train_data/test_seg/

Parameters type: cli arguments

Execute:

python test_segment.py -m train_data/models/vgg7/pucpr-200.model -d train_data/test_seg/

Related files:

• detect_contour.py
• train_data/test_img/

Parameters type: built-in

Execute:

python detect_contour.py

1. Prepare

Related files:

• data_generator.py
• src_img/
• train_data/train/

Parameters type: built-in

Execute:

python data_generator.py

2. Train

Related files:

• train_network.py
• train_data/train/
• train_data/models/

Parameters type: cli arguments

Execute:

# All datasets
python train_network.py -i train_data/train/ -o train_data/models/vgg7/all

# Specific dataset
python train_network.py -i train_data/train/pucpr -o train_data/models/vgg7/pucpr

3. Assess

Related files:

• test_image.py
• calc_accuracy.py
• train_data/models/
• train_data/test_img/

Parameters type: built-in

Execute:

# Test image
python test_image.py

# Calc accuracy
python calc_accuracy.py

Accuracy

How to decide training size:

PUCPR 36 days, UFPR04 31 days, UFPR05 33 days

About 12 hours (6am - 6pm), 12 sheets per hour (Take one shot every 5 minutes)

12x12x0.2=28.8, 33x0.2=6.6, 28.8x6.6=190.08

Parameters:

Testing size: 20,000 parking spaces on each parking lot

Training images: 200

Epochs: 5

Parking lot	Network	Acc (PUCPR)	Acc (UFPR04)	Acc (UFPR05)
PUCPR	AlexNet	99.87%	98.06%	94.69%
PUCPR	LeNet	99.84%	97.56%	84.78%
PUCPR	VGG-7	99.95%	99.02%	95.77%
PUCPR	VGG-13	99.93%	94.28%	92.81%
PUCPR	VGG-13V	99.97%	98.37%	95.13%
PUCPR	VGG-16	99.77%	96.79%	92.13%
ALL	AlexNet	99.44%	93.01%	98.52%
ALL	LeNet	99.58%	97.27%	98.83%
ALL	VGG-7	99.75%	98.80%	99.57%
ALL	VGG-13V	99.67%	98.79%	99.07%
ALL	VGG-16	93.20%	77.69%	98.87%

Epochs: 5

Parking lot	Network	Images	Acc (PUCPR)	Acc (UFPR04)	Acc (UFPR05)
PUCPR	VGG-7	100	89.72%	49.41%	88.92%
PUCPR	VGG-7	200	99.95%	99.02%	95.77%
PUCPR	VGG-7	1000	99.99%	98.17%	96.25%

Training images: 200

Parking lot	Network	Epochs	Acc (PUCPR)	Acc (UFPR04)	Acc (UFPR05)
ALL	VGG-16	5	93.20%	77.69%	98.87%
ALL	VGG-16	10	98.08%	92.81%	96.76%
ALL	VGG-16	15	97.48%	95.17%	98.05%

Training images: 800

Parking lot	Network	Epochs	Acc (PUCPR)	Acc (UFPR04)	ACC (UFPR05)
ALL	VGG-7	5	99.94%	99.09%	99.44%
ALL	VGG-7	8	99.69%	98.47%	99.30%
ALL	VGG-7	12	99.99%	99.72%	99.52%

Parking lot	Network	Epochs	Acc (PUCPR)	Acc (UFPR04)	ACC (UFPR05)
ALL	VGG-13V	5
ALL	VGG-13V	10
ALL	VGG-13V	20	99.95%	99.63%	99.38%

Regularization

Use PUCPR training set & VGG-7 network

Id	Method	Acc (PUCPR)	Acc (UFPR04)	Acc (UFPR05)
r0	None	99.81%	92.55%	92.53%
r1	Batch Normalization	99.94%	96.73%	91.02%
r2	L2 regularization	99.43%	91.27%	77.62%
r3	Dropout	99.95%	96.95%	92.84%
r4	BN + Dropout	99.95%	99.02%	95.77%
r5	BN + L2 + Dropout	99.87%	92.51%	94.45%

Use PUCPR training set & VGG-13V network

Id	Method	Acc (PUCPR)	Acc (UFPR04)	Acc (UFPR05)
r0	None	99.97%	98.69%	89.49%
r1	Batch Normalization	99.94%	97.48%	96.37%
r2	L2 regularization	99.91%	96.14%	93.39%
r3	Dropout	55.12%	55.67%	42.29%
r4	BN + Dropout	99.82%	94.05%	93.66%
r5	BN + L2 + Dropout	99.97%	98.37%	95.13%

Verbose

train_network.py

(Tensorflow-3.6) C:\Users\AWAKE\Desktop\GP\ips>python train_network.py -d train_data/train/ -m train_data/models/vgg16-200.model
Using TensorFlow backend.
[INFO] loading images...
[INFO] compiling model...
2019-03-30 14:43:39.127451: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1432] Found device 0 with properties:
name: GeForce GTX 1060 6GB major: 6 minor: 1 memoryClockRate(GHz): 1.7845
pciBusID: 0000:01:00.0
totalMemory: 6.00GiB freeMemory: 4.96GiB
2019-03-30 14:43:39.135581: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1511] Adding visible gpu devices: 0
2019-03-30 14:43:40.248390: I tensorflow/core/common_runtime/gpu/gpu_device.cc:982] Device interconnect StreamExecutor with strength 1 edge matrix:
2019-03-30 14:43:40.253854: I tensorflow/core/common_runtime/gpu/gpu_device.cc:988]      0
2019-03-30 14:43:40.256276: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1001] 0:   N
2019-03-30 14:43:40.260234: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 4714 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1060 6GB, pci bus id: 0000:01:00.0, compute capability: 6.1)
[INFO] training network...
Epoch 1/5
819/819 [==============================] - 95s 116ms/step - loss: 0.2855 - acc: 0.9539 - val_loss: 0.1071 - val_acc: 0.9681
Epoch 2/5
819/819 [==============================] - 87s 107ms/step - loss: 0.0765 - acc: 0.9789 - val_loss: 7.0363 - val_acc: 0.5226
Epoch 3/5
819/819 [==============================] - 87s 107ms/step - loss: 0.1046 - acc: 0.9717 - val_loss: 0.0206 - val_acc: 0.9933
Epoch 4/5
819/819 [==============================] - 89s 108ms/step - loss: 0.0598 - acc: 0.9838 - val_loss: 0.0252 - val_acc: 0.9931
Epoch 5/5
819/819 [==============================] - 94s 115ms/step - loss: 0.0407 - acc: 0.9885 - val_loss: 0.2214 - val_acc: 0.9199
[INFO] serializing network...

calc_accuracy.py

(Tensorflow-3.6) C:\Users\AWAKE\Desktop\GP\ips>python calc_accuracy.py
Using TensorFlow backend.
[INFO] loading network...
2019-03-31 20:00:34.046661: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1432] Found device 0 with properties:
name: GeForce GTX 1060 6GB major: 6 minor: 1 memoryClockRate(GHz): 1.7845
pciBusID: 0000:01:00.0
totalMemory: 6.00GiB freeMemory: 4.96GiB
2019-03-31 20:00:34.055607: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1511] Adding visible gpu devices: 0
2019-03-31 20:00:35.094689: I tensorflow/core/common_runtime/gpu/gpu_device.cc:982] Device interconnect StreamExecutor with strength 1 edge matrix:
2019-03-31 20:00:35.100110: I tensorflow/core/common_runtime/gpu/gpu_device.cc:988]      0
2019-03-31 20:00:35.103378: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1001] 0:   N
2019-03-31 20:00:35.107615: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 4714 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1060 6GB, pci bus id: 0000:01:00.0, compute capability: 6.1)
100%|#######################################################################################################| 60000/60000 [05:02<00:00, 198.42it/s]
Accuracy of PUCPR: 98.87%
Accuracy of UFPR04: 95.65%
Accuracy of UFPR05: 86.81%

References

• Python 风格指南
• Amazon Machine Learning