SUTD 50.035 Computer Vision Course Project
Security object detection
Model has been trained to look for Handgun,Knife,Beer,Baseball bat,Skateboard
*YOLO Variant was just a test and not implemented in the end, therefore has no training or evaluation method
For SSD, to train the model,
Ensure that data is in the form of
.
+-- open_images/
| +-- test/
| +-- <testimages>
| +-- train/
| +-- <trainimages>
| +-- validation/
| +-- <validationimages>
| +-- class-descriptions-boxable.csv
| +-- sub-test-annotations-bbox.csv
| +-- sub-train-annotations-bbox.csv
| +-- sub-validation-annotations-bbox.csv
| +-- test-annotations-bbox.csv
| +-- train-annotations-bbox.csv
| +-- validation-annotations-bbox.csv
You may use the given script in the file to download it (Script from https://github.com/qfgaohao/pytorch-ssd). and example on how to run it is as such: python open_images_downloader.py --root ./data/open_images --class_names "Handgun,Knife,Beer,Baseball bat,Skateboard"
Edit the hyper parameters in train.py and run the file
Next, to eval, run eval.py (similarly after changing hyperparmeters)
Finally, you can draw boundingboxes on your eval using draweval.py, be sure to edit the respective file path
Requires the following
$ pip install python-opencv
$ pip install torch
$ pip install torchvision
$ pip install numpy
To run the livefeed with detection
$ git clone https://github.com/scarmaten/comvision.git
$ cd comvision/securitybyyolo
$ python livedarknet.py
Press q to quit
Requires the following
$ pip install python-opencv
$ pip install torch
$ pip install torchvision
$ pip install numpy
To download our model, head over to https://www.dropbox.com/s/lv5ye9u0k7zpjvy/vgg-Epoch-10-Loss-3.0180892944335938.pth?dl=0
To download our version 2 model (using kernel size 1 and 0 padding on last layer), head over to https://www.dropbox.com/s/9j7mf9tvi01i10j/v2-Epoch-39-Loss-7.86013218334743.pth?dl=0
Version 2 works better from our testing
Version 1
self.classification_headers = ModuleList([
Conv2d(in_channels=512, out_channels=4 * num_classes, kernel_size=3, padding=1),
Conv2d(in_channels=1024, out_channels=6 * num_classes, kernel_size=3, padding=1),
Conv2d(in_channels=512, out_channels=6 * num_classes, kernel_size=3, padding=1),
Conv2d(in_channels=256, out_channels=6 * num_classes, kernel_size=3, padding=1),
Conv2d(in_channels=256, out_channels=4 * num_classes, kernel_size=3, padding=1),
Conv2d(in_channels=256, out_channels=4 * num_classes, kernel_size=3, padding=1),
])
self.regression_headers = ModuleList([
Conv2d(in_channels=512, out_channels=4 * 4, kernel_size=3, padding=1),
Conv2d(in_channels=1024, out_channels=6 * 4, kernel_size=3, padding=1),
Conv2d(in_channels=512, out_channels=6 * 4, kernel_size=3, padding=1),
Conv2d(in_channels=256, out_channels=6 * 4, kernel_size=3, padding=1),
Conv2d(in_channels=256, out_channels=4 * 4, kernel_size=3, padding=1),
Conv2d(in_channels=256, out_channels=4 * 4, kernel_size=3, padding=1),
])
Version 2
self.classification_headers = ModuleList([
Conv2d(in_channels=512, out_channels=4 * num_classes, kernel_size=3, padding=1),
Conv2d(in_channels=1024, out_channels=6 * num_classes, kernel_size=3, padding=1),
Conv2d(in_channels=512, out_channels=6 * num_classes, kernel_size=3, padding=1),
Conv2d(in_channels=256, out_channels=6 * num_classes, kernel_size=3, padding=1),
Conv2d(in_channels=256, out_channels=4 * num_classes, kernel_size=3, padding=1),
Conv2d(in_channels=256, out_channels=4 * num_classes, kernel_size=1, padding=0),
])
self.regression_headers = ModuleList([
Conv2d(in_channels=512, out_channels=4 * 4, kernel_size=3, padding=1),
Conv2d(in_channels=1024, out_channels=6 * 4, kernel_size=3, padding=1),
Conv2d(in_channels=512, out_channels=6 * 4, kernel_size=3, padding=1),
Conv2d(in_channels=256, out_channels=6 * 4, kernel_size=3, padding=1),
Conv2d(in_channels=256, out_channels=4 * 4, kernel_size=3, padding=1),
Conv2d(in_channels=256, out_channels=4 * 4, kernel_size=1, padding=0),
])
$ git clone https://github.com/scarmaten/comvision.git
$ cd comvision/securitybyssd
$ python livedetection.py
Press q to quit
- Kah Wee
- Rachel
- Venessa
- Jun Qing
This project referenced and adapted various sources
- https://blog.paperspace.com/how-to-implement-a-yolo-object-detector-in-pytorch/
- https://github.com/qfgaohao/pytorch-ssd
- https://github.com/amdegroot/ssd.pytorch
Dataset used for training from OpenImages