CVEET

Classifying Vehicles Entering and Exiting a tunnel, MSc thesis.

General Overview

• annotations - Object Detection Train & Test binary files + Label files.
• exported-models - Saved Models.
• fig - Saved Figures, Plots, etc.
• images - Image data and annotation data.
• logs - Various logs for profiling, evaluation, etc.
• models - Object Detection model configurations.
• pre-trained-models - Object Detection pre-trained models.
• report - Presentation and pdf report.
• src - Source Code.

Code Overview

• preprocessing.ipynb - Used for splitting the data into train and test, creates a file which holds the filenames rather than copying over the same files. Additionally it includes some histogram plots of the various vehicle distributions, and an algorithm for extracting the objects in the annotated images, such that they can be used in a regular CNN.
• gen-tf-record.ipynb - Generates the TF Record files from the train and test image folders. These folders contain the mapping files which hold images and their respective annotation file paths.
• convert-data-labels.ipynb - Is used to convert other annotation formats, and image formats to the correct formats. Various parsers and translation stuff logic is found here.
• webcam-scraper [.ipynb, .py] - Is used for fetching images from certain webcams.
• utils.py - Utility functions for plotting and parsing XML.
• stream.py - Realtime R-CNN / SSD based object detection system, utilizes PyQt5 for the GUI.
• object-detection.ipynb - Used for testing the R-CNN / SSD approach on single images.
• object-detection-bgsub.ipynb - Used for testing vanilla CNN with BG subtraction logic.
• detection.py - Contains object detection utility functions for the R-CNN / SSD approach.
• eval-stat.py - Used for creating plots for the evaluation of the custom data generator.
• transfer-learning.py - Used for training vanilla CNN, doing hyperparam tuning, testing distributions, etc.
• testing.ipynb - Misc testing, and various plotting.
• profiling.py - Used for profiling the CNN and region proposal method.

Executing Training

Run train.sh for object-detection or train-cnn for a general CNN. Export object-detection model with exportmdl.sh.

Prerequisites (pip packages)

• OpenCV
• Tensorflow
• Tensorflow Hub
• Tensorflow Object Detection API
• NumPy
• matplotlib
• scipy
• sklearn
• PyQt5
• nvidia_smi
• psutil

Scoreboard

Feature Vector	Accuracy	Epochs	Batch Size	Optimizer	Learning Rate	Dropout	Label Smoothing	Regularization	Generator
MobileNET + Tuning	0.949	30	64	RMSProp	0.001	0.1	0.15	L1 0.001	Custom - Distr [0.60, 0.70, 0.25, 0.35, 0.65]
EfficientNET b7	0.930	50	32	RMSProp	0.001	0.25	0.15	L1 0.001	Custom
MobileNET	0.927	50	32	Adam	0.001	0.2	0.2	L1 0.001	Custom
MobileNET	0.922	65	32	RMSProp	0.001	0.2	0.15	L1 0.001	Custom
MobileNET	0.900	65	32	Adam	0.01	0.3	0.2	L1 0.001	Custom
MobileNET	0.940	100	16	SGD	0.001	0.1	0.1	L1 0.001	Less Data
MobileNET	0.810	80	16	SGD	0.001	0.0	0.0	L1 0.001	Default
InceptionNET	0.760	4	16	RMSProp	0.0001	0.1	0.1	L2 0.0001	Default

R-CNN Realtime Demo

Data

• CVEET Data
• KITTI Data
• ExDark Data

Misc

• CVEET Exported Models & Pre-Trained Models
• Check the report folder for a presentation of the work and pdf of the thesis.