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

Easy to use Python camera interface for NVIDIA Jetson

A fast implementation of bss_eval metrics for blind source separation

The project was to detect traffic signs, based on the Megengine framework.

Robust Self-augmentation for NER with Meta-reweighting

Rendering Point Clouds with Compute Shaders

✂️ EyeLipCropper is a Python tool to crop eyes and mouth ROIs of the given video.

Transfer Learning Shootout for PyTorch's model zoo (torchvision)

OpenABC-D: A Large-Scale Dataset For Machine Learning Guided Integrated Circuit Synthesis

FCA: Learning a 3D Full-coverage Vehicle Camouflage for Multi-view Physical Adversarial Attack

Easily Process a Batch of Cox Models

Official code repository for the work: "The Implicit Values of A Good Hand Shake: Handheld Multi-Frame Neural Depth Refinement"

Fine-grained Post-training for Improving Retrieval-based Dialogue Systems - NAACL 2021

Non-Homogeneous Poisson Process Intensity Modeling and Estimation using Measure Transport

Posterior temperature optimized Bayesian models for inverse problems in medical imaging

Class-Attentive Diffusion Network for Semi-Supervised Classification [AAAI'21] (official implementation)

UI2I via StyleGAN2 - Unsupervised image-to-image translation method via pre-trained StyleGAN2 network

Python Library for learning (Structure and Parameter) and inference (Statistical and Causal) in Bayesian Networks.

The Python ensemble sampling toolkit for affine-invariant MCMC

AutoVideo: An Automated Video Action Recognition System

Transformer in Computer Vision