All Projects → shaoshengsong → MobileNetV3-SSD-Compact-Version

shaoshengsong / MobileNetV3-SSD-Compact-Version

Licence: other
MobileNetV3 SSD的简洁版本

Programming Languages

python
139335 projects - #7 most used programming language

Projects that are alternatives of or similar to MobileNetV3-SSD-Compact-Version

MobileNet-SSD-windows
No description or website provided.
Stars: ✭ 91 (+40%)
Mutual labels:  ssd, mobilenet, mobilenet-ssd
Mobilenet Ssd
Caffe implementation of Google MobileNet SSD detection network, with pretrained weights on VOC0712 and mAP=0.727.
Stars: ✭ 1,805 (+2676.92%)
Mutual labels:  ssd, mobilenet, mobilenet-ssd
hand detection
A Light CNN based Method for Hand Detection and Orientation Estimation
Stars: ✭ 116 (+78.46%)
Mutual labels:  ssd, mobilenet, mobilenet-ssd
MobileNetV3-TF
Tensorflow implementation for two new MobileNetV3 models!
Stars: ✭ 25 (-61.54%)
Mutual labels:  mobilenet, mobilenetv3, mobilenetv3-large
pedestrian recognition
A simple human recognition api for re-ID usage, power by paper https://arxiv.org/abs/1703.07737
Stars: ✭ 29 (-55.38%)
Mutual labels:  mobilenet, mobilenet-ssd
TensorMONK
A collection of deep learning models (PyTorch implemtation)
Stars: ✭ 21 (-67.69%)
Mutual labels:  ssd-mobilenet, mobilenet-ssd
Mobilenetv3 Ssd
MobileNetV3-SSD for object detection and implementation in PyTorch
Stars: ✭ 188 (+189.23%)
Mutual labels:  ssd, mobilenet
ModelZoo.pytorch
Hands on Imagenet training. Unofficial ModelZoo project on Pytorch. MobileNetV3 Top1 75.64🌟 GhostNet1.3x 75.78🌟
Stars: ✭ 42 (-35.38%)
Mutual labels:  mobilenet, mobilenetv3
PaddlePaddle-SSD
基于PaddlePaddle实现的SSD,包括MobileNetSSD,MobileNetV2SSD,VGGSSD,ResNetSSD
Stars: ✭ 47 (-27.69%)
Mutual labels:  ssd, ssd-mobilenet
Ssds.pytorch
Repository for Single Shot MultiBox Detector and its variants, implemented with pytorch, python3.
Stars: ✭ 570 (+776.92%)
Mutual labels:  ssd, mobilenet
Mobilenetv2 Ssdlite
Caffe implementation of SSD and SSDLite detection on MobileNetv2, converted from tensorflow.
Stars: ✭ 435 (+569.23%)
Mutual labels:  ssd, mobilenet
Tensorflow Face Detection
A mobilenet SSD based face detector, powered by tensorflow object detection api, trained by WIDERFACE dataset.
Stars: ✭ 711 (+993.85%)
Mutual labels:  ssd, mobilenet
Tf Object Detection
Simpler app for tensorflow object detection API
Stars: ✭ 91 (+40%)
Mutual labels:  ssd, mobilenet
Ssds pytorch
Multiple basenet MobileNet v1,v2, ResNet combined with SSD detection method and it's variants such as RFB, FSSD etc.
Stars: ✭ 71 (+9.23%)
Mutual labels:  ssd, mobilenet
Proctoring Ai
Creating a software for automatic monitoring in online proctoring
Stars: ✭ 155 (+138.46%)
Mutual labels:  ssd, mobilenet
Ssd keras
简明 SSD 目标检测模型 keras version(交通标志识别 训练部分见 dev 分支)
Stars: ✭ 152 (+133.85%)
Mutual labels:  ssd
Traffic Sign Detection
Traffic Sign Detection. Code for the paper entitled "Evaluation of deep neural networks for traffic sign detection systems".
Stars: ✭ 200 (+207.69%)
Mutual labels:  ssd
Kdiskmark
A simple open-source disk benchmark tool for Linux distros
Stars: ✭ 152 (+133.85%)
Mutual labels:  ssd
Mmdetection
OpenMMLab Detection Toolbox and Benchmark
Stars: ✭ 17,646 (+27047.69%)
Mutual labels:  ssd
Ssd keras
A Keras port of Single Shot MultiBox Detector
Stars: ✭ 1,763 (+2612.31%)
Mutual labels:  ssd

MobileNetV3-SSD-Compact-Version

MobileNetV3 SSD的简洁版本

希望尝试新技术请到这里 https://github.com/shaoshengsong/quarkdet 一个轻量级的目标检测包括多种模型

环境 Ubuntu18.04 版本 PyTorch 1.4

如果您想从头开始需要使用 mbv3_large.old.pth.tar 是backbone用来参数初始化的模型 有可能会历经坎坷

简便方式就是使用预训练模型 如果您要直接测试,模型改名为checkpoint_ssd300.pth.tar 模型下载地址 链接:https://pan.baidu.com/s/1BZz9X7w3xaopf1dKLkyMcA 提取码:gwwv

模型测试结果 mAP 0.679 (未在COCO数据集做预训练版本)

使用步骤 一 下载VOC数据集之后,将VOCtrainval_06-Nov-2007和VOCtest_06-Nov-2007合并在一起 数据集下载 可以看这里 https://blog.csdn.net/flyfish1986/article/details/95367745

二 先打开create_data_lists.py文件 改成自己数据集的路径

三 如果想使用mobilenetv3的预训练模型,打开mode.py 找到 def init_weights(self, pretrained=None):#"./mbv3_large.old.pth.tar" 替换成 def init_weights(self, pretrained="./mbv3_large.old.pth.tar"):

四 运行训练命令python train.py

五 测试mAP命令 python eval.py

以下是站在巨人肩膀上的那个人的肩膀上 hard negative mining和L2Norm没有封装,在代码里直白的编写

mobilenetv3的预训练模型从这里下载

mmdetection版本测试结果没有达到官方所说的水平,而是mAP没有到0.5

增强函数改编自

计算mAP的博客

坐标变换部分来自

关于提高mAP的做法

如果还想提高mAP怎么办

本repo没有在COCO数据集下训练,只训练了VOC0712数据集 如果想得到更高的mAP,可以尝试下先在COCO数据集下训练,这时候得到的模型作为预训练模型,然后在VOC0712数据集中微调,最后再在VOC2007下测试,看看是不是会得到大于0.619mAP的数

我使用第二版的部分源码做的PyTorch版的VGG-SSD,测试不同的学习率 如果按照常规做法,在训练达到某个epoch 原学习率* 0.1 , 在训练达到某个epoch 原学习率*0.01这样最终结果会得到比原论文还优秀的0.77x mAP,x代表一个很大的个位数. 如果单独采用余弦退火,mAP可以到0.771mAP,低了0.00y. 所以想增加mAP可以尝试下常规做法的学习率.

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].