v1.0.4

• Add pre-processing and post-processing to ONNX in WHENet

  • Pre-processing adds normalization processing
  • Post-processing adds conversion process to angles (-180° <= 0° <= 180°)
    
    

• disable_front_side_detection
  
  

v1.0.3

1. add script for initializing batch size
2. add ONNX with variable batch size

• ONNX (1 batch, 4 batch, 8 batch, 16 batch, N batch)
• TensorFlow saved_model v2
• TFLite (Float32/Float16/INT8/Dynamic Range Quant)
• CoreML
• EdgeTPU
• TensorRT
• TFJS
• YOLOv4 / YOLOv4-tiny-3L Head Detection

v1.0.2

• ONNX (1batch, 4batch, 8batch, 16batch)
• TensorFlow saved_model v2
• TFLite (Float32/Float16/INT8/Dynamic Range Quant)
• CoreML
• EdgeTPU
• TensorRT
• TFJS
• YOLOv4 / YOLOv4-tiny-3L Head Detection

v1.0.1

• ONNX
• TensorFlow saved_model v2
• TFLite (Float32/Float16/INT8/Dynamic Range Quant)
• CoreML
• EdgeTPU
• TensorRT
• TFJS
• YOLOv4 / YOLOv4-tiny-3L Head Detection

v1.0.0

• ONNX
• TensorFlow saved_model v2
• TFLite (Float32/Float16/INT8/Dynamic Range Quant)
• CoreML
• EdgeTPU
• TensorRT
• TFJS