White Blood Cell Detection and Classification using Single Shot Multibox Detector and Deployment

---

This is a PyTorch Tutorial to Object Detection.

This is blatanly "borrowed" from sgrvinod's great tutorial

Basic knowledge of PyTorch, convolutional neural networks is assumed.

---

Set up

• Run command

git clone https://github.com/tranduchuy682/SSD.git
pip install -r requirements.txt

• Or run .sh file

---

Get dataset

• The dataset can be downloaded or cloned here
• Format

├── AllDatabase
|  ├── BCCD Database
|  ├── LISCDatabase
|  |    └── Ground Truth Segmentation
|  |    └── Main Dataset
|  |    └── More Dataset without Ground Truth/alll
|  ├── RaabinDatabase
|  |    └── GrTh
|  |    └── TestA
|  |    └── TestB/Test-B
|  |    └── Train
|  ├── bbox.csv
|  ├── test_bbox.csv
|  └── train_bbox.csv
├── train.py
├── utils.py
...

• Run

cd ssd
git clone https://github.com/tranduchuy682/AllDatabase.git

---

Prepare the dataset for training and evaluate

• Run command

python3 create_data_lists.py

---

Training

• Backbones: There are 3 backbone have been used

• VGG16 backbone

  

• Resnet18 backbone

• We have modified our resnet18 backbone to get resnet18* - more suitable with our dataset

  

• MobileNetV3 backbone

  

• Config

backbone == ["resnet18","mobilenetv3","vgg16"]
epoch = 150 or any int value

• Run command

python3 train.py resnet18 150

---

Testing

• Run command

python3 eval.py

Inference

• Run command

python3 detect.py

Deploy

• Run command

python3 app.py

• Demo