Non-uniform Step Size Quantization for Accurate Post-training Quantization.
Sangyun OhHyeonuk SimJounghyun KimJongeun LeePublished in: ECCV (11) (2022)
Keyphrases
- step size
- quantization step
- stochastic gradient descent
- convergence rate
- cost function
- convergence speed
- dct domain
- steepest descent method
- hessian matrix
- faster convergence
- training algorithm
- high quality
- evolutionary programming
- matrix factorization
- filter bank
- frequency domain
- jpeg compression
- image quality
- image segmentation
- computer vision
- neural network