YOLOv3 via a MobileNetV3 backbone for text detection; pruned, quantized, optimized, and explained for deployment on mobile devices. Primarily intended as a single source for learning about YOLO(v3) in an applied manner.
- Pretrained MobileNetV2 backbone
- Introduce the YOLOv3 paradigm
- Basic Pruning, Quantization integration
- Training pipeline (for ICDAR 2015)
- Switch backbone to MobileNetV3
- Mixed Precision Training
- Pruning and quantization
- Add textbook-style explanations for YOLOv3
- Optimize, expand to applicability to other datasets
- Yolov3: An incremental improvement [Farhadi, A. & Redmon, J., 2018]
- ICDAR 2015 Dataset Kaggle.com
- Mobile data science and intelligent apps: concepts, AI-based modeling and research directions [Sarker, et al. 2021]
- Faster R-CNN: Towards real-time object detection with region proposal networks [Ren, et al. 2016]
- Histograms of oriented gradients for human detection [Dalal, N., & Triggs, B. 2005]
- Distance-IoU loss: Faster and better learning for bounding box regression [Zheng, et al. 2020]
- Focal loss for dense object detection [Ross, et al. 2017]
- Does label smoothing mitigate label noise? [Lukasik, et al. 2020]
- Searching for activation functions [Ramachandran, et al. 2017]
- Searching for mobilenetv3 [Howard, et al. 2019]
- ECA-Net: Efficient channel attention for deep convolutional neural networks [Wang, et al. 2020]
- Xception: Deep learning with depthwise separable convolutions [Chollet, F. 2017]
- Dropout: a simple way to prevent neural networks from overfitting [Srivastava, et al. 2014]
- MixUp: Beyond empirical risk minimization [Zhang, H. 2017]
- Super-convergence: Very fast training of neural networks using large learning rates [Smith, L. N., & Topin, N. 2019]
- Lookahead optimizer: k steps forward, 1 step back [Zhang, et al. 2019]
- Methods for pruning deep neural networks [Vadera, S., & Ameen, S. 2022]
- Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding [Han, et al. 2015]