YOLO v4 and TF 2.0

Hi everyone,

               Welcome to my second course on computer vision. In this course, you will understand the two most latest State Of The Art(SOTA) object detection architecture, which is YOLOv4 and TensorFlow 2.0 and its training pipeline. I also included a one-time labeling strategy, so that you won't have to re-label the image for TensorFlow training. The course is split into 9 parts.

1. Anaconda installation.

2. Image dataset resizing.

3. Image dataset labeling.

4. YOLO to PASCAL VOC conversion for TF2.0 training.

5. YOLOv4 training and tflite conversion on Google Colab.

6. YOLOv4 Android deployment.

7. SSD Mobilenet TF2.0 training and tflite conversion on Google Colab.

8. SSD Mobilenet Android deployment.

9. YOLOv4 and SSD technical details. Which include

   Basics

   1. Precision  and Recall

   2. IoU(Intersection Over Union)

   3. Mean Average Precision/Average Precision(mAP/AP)

   4. Batch Normalization

   5. Residual blocks

   6. Activation function

   7. Max pooling

   8. Feature Pyramid Networks(FPN)

   9. Path Aggregation Network (PAN)

   10. SPP (spatial pyramid pooling layer)

   11. Channel Attention Module(CAM) and Spatial Attention Module (SAM)

   YOLOv4 - Technical details

   1. Backbone

   2. Cross-Stage-Partial-connections (CSP)

   3. YOLO with SPP

   4. PAN in YOLOv4

   5. Spatial Attention Module (SAM) in YOLOv4

   6. Bag of freebies (Bof) and Bag of specials (BoS)

   SSD - Technical details

   1. Architecture overview and working

   2. Loss functions

   YOLO vs SSD

   1. Speed and accuracy benchmarking