yolov5目标检测模型的知识蒸馏（基于响应的蒸馏）

Last update: Dec 04, 2022

Overview

代码地址：

https://github.com/Sharpiless/yolov5-knowledge-distillation

教师模型：

python train.py --weights weights/yolov5m.pt \
        --cfg models/yolov5m.yaml --data data/voc.yaml --epochs 50 \
        --batch-size 8 --device 0 --hyp data/hyp.scratch.yaml

蒸馏训练：

python train.py --weights weights/yolov5s.pt \
        --cfg models/yolov5s.yaml --data data/voc.yaml --epochs 50 \
        --batch-size 8 --device 0 --hyp data/hyp.scratch.yaml \
        --t_weights yolov5m.pt --distill

训练参数:

--weights：预训练模型

--t_weights：教师模型权重

--distill：使用知识蒸馏进行训练

--dist_loss：l2或者kl

--temperature：使用知识蒸馏时的温度

使用《Object detection at 200 Frames Per Second》中的损失

这篇文章分别对这几个损失函数做出改进，具体思路为只有当teacher network的objectness value高时，才学习bounding box坐标和class probabilities。

实验结果：

这里假设VOC2012中新增加的数据为无标签数据（2k张）。

教师模型	训练方法	蒸馏损失	P	R	mAP50
无	正常训练	不使用	0.7756	0.7115	0.7609
Yolov5l	output based	l2	0.7585	0.7198	0.7644
Yolov5l	output based	KL	0.7417	0.7207	0.7536
Yolov5m	output based	l2	0.7682	0.7436	0.7976
Yolov5m	output based	KL	0.7731	0.7313	0.7931

参数和细节正在完善，支持KL散度、L2 logits损失和Sigmoid蒸馏损失等

1. 正常训练：

2. L2蒸馏损失：

我的公众号：

关于作者

B站：https://space.bilibili.com/470550823

CSDN：https://blog.csdn.net/weixin_44936889

AI Studio：https://aistudio.baidu.com/aistudio/personalcenter/thirdview/67156

Github：https://github.com/Sharpiless

yolov5目标检测模型的知识蒸馏（基于响应的蒸馏）

Related tags

Overview

代码地址：

教师模型：

蒸馏训练：

训练参数:

实验结果：

1. 正常训练：

2. L2蒸馏损失：

我的公众号：

关于作者

Owner

This repository provides an efficient PyTorch-based library for training deep models.

DeepCO3: Deep Instance Co-segmentation by Co-peak Search and Co-saliency

Python package to add text to images, textures and different backgrounds

Improving Transferability of Representations via Augmentation-Aware Self-Supervision

Kinetics-Data-Preprocessing

This is the code for the paper "Motion-Focused Contrastive Learning of Video Representations" (ICCV'21).

Merlion: A Machine Learning Framework for Time Series Intelligence

Siamese-nn-semantic-text-similarity - A repository containing comprehensive Neural Networks based PyTorch implementations for the semantic text similarity task

Performance Analysis of Multi-user NOMA Wireless-Powered mMTC Networks: A Stochastic Geometry Approach

Spatial Single-Cell Analysis Toolkit

Deep Distributed Control of Port-Hamiltonian Systems

Boston House Prediction Valuation Tool

The tl;dr on a few notable transformer/language model papers + other papers (alignment, memorization, etc).

PyBullet CartPole and Quadrotor environments—with CasADi symbolic a priori dynamics—for learning-based control and reinforcement learning

Speech Separation Using an Asynchronous Fully Recurrent Convolutional Neural Network

RetinaFace: Deep Face Detection Library in TensorFlow for Python

Matlab Python Heuristic Battery Opt - SMOP conversion and manual conversion

Official Pytorch implementation for 2021 ICCV paper "Learning Motion Priors for 4D Human Body Capture in 3D Scenes" and trained models / data

Implementation of character based convolutional neural network

Shuffle Attention for MobileNetV3