objdet

object detection algorithms

这个仓库旨在实现常用的目标检测算法，主要参考如下：

• mmdetection，参考CUHK检测框架的思路。
• ssd.pytorch
• Object-Detection
• 2015-10-09-object-detection.md handong收集的相关目标检测论文目录；
• awesome-object-detection，awesome系列，参考合并了handong相关目录；
• 目标检测 Object Detection 博客整理收集的相关资料；
• Review of Deep Learning Algorithms for Object Detection 相关目标检测DL算法综述；
• deep_learning_object_detection，其中增加了可视化paper表格，非常直观；
• Deep Learning for Generic Object Detection: A Survey，深度学习通用目标检测调研；
• ...

图片来自于deep_learning_object_detection

论文资料

• DeNet: Scalable Real-time Object Detection with Directed Sparse Sampling 相关代码denet。
• Soft Proposal Networks for Weakly Supervised Object Localization 相关代码SPN.pytorch
• ICCV 2015 Tutorial on Tools for Efficient Object Detection ICCV 2015中举办的关于目标检测的教程，可以参考。
• Deep Learning for Objects and Scenes CVPR 2017关于目标检测的教程。
• RSA-for-object-detection-cpp-version RSA-for-object-detection 相关论文Recurrent Scale Approximation for Object Detection in CNN
• DetNet: A Backbone network for Object Detection
• 小目标检测，参考如下
• 遮挡目标检测，参考如下
• 视频目标检测，参考video_obj
• domain目标检测，参考如下
• 非极大值处理，参考如下
• 弱监督目标检测，参考如下
• 困难样本采样策略，参考如下
• 文本检测，参考如下
• 类别不平衡目标检测，参考如下
• 小数据集目标检测，参考如下
• A unified multi-scale deep convolutional neural network for fast object detection
• How Far are We from Solving Pedestrian Detection? 行人检测
• Taking a Deeper Look at Pedestrians 行人检测
• Integralchannel features 行人检测
• Fast Feature Pyramids for Object Detection 行人检测
• What Can Help Pedestrian Detection? 行人检测
• Citypersons: A di- verse dataset for pedestrian detection 行人检测数据集
• DenseBox: Unifying Landmark Localization with End to End Object Detection，不使用anchor的检测方法
• UnitBox: An Advanced Object Detection Network
• Discriminative models for multi-class object layout
• Learning Transferable Architectures for Scalable Image Recognition，自学网络结构；
• Single-Shot Refinement Neural Network for Object Detection，S3FD和RefineDet论文都是同一个作者；
• HyperNet: Towards Accurate Region Proposal Generation and Joint Object Detection
• Hypercolumns for Object Segmentation and Fine-grained Localization

小目标检测

• Feature-Fused SSD: Fast Detection for Small Objects
• Perceptual Generative Adversarial Networks for Small Object Detection，GAN小目标检测，暂且不看；
• Detecting and counting tiny faces
• Seeing Small Faces from Robust Anchor's Perspective
• Face-MagNet: Magnifying Feature Maps to Detect Small Faces
• Small-scale Pedestrian Detection Based on Somatic Topology Localization and Temporal Feature Aggregation
• MDSSD: Multi-scale Deconvolutional Single Shot Detector for Small Objects
• CMS-RCNN: Contextual Multi-Scale Region-based CNN for Unconstrained Face Detection 集成人体上下文信息来帮助推理人脸位置；
• Finding Tiny Faces，多级图像金字塔进行multi-scale训练和测试；
• S3FD: Single Shot Scale-invariant Face Detector single shot尺度等变的人脸检测器；

遮挡目标检测

• Repulsion Loss: Detecting Pedestrians in a Crowd

domain目标检测

• Cross-Domain Weakly-Supervised Object Detection through Progressive Domain Adaptation

文本检测

• Deep Direct Regression for Multi-Oriented Scene Text Detection
• TextBoxes: A Fast Text Detector with a Single Deep Neural Network
• Detecting Text in Natural Image with Connectionist Text Proposal Network
• R2CNN: Rotational Region CNN for Orientation Robust Scene Text Detection
• EAST: An Efficient and Accurate Scene Text Detector
• Detecting Oriented Text in Natural Images by Linking Segments
• Arbitrary-Oriented Scene Text Detection via Rotation Proposals
• Scene Text Detection via Holistic, Multi-Channel Prediction
• Deep Matching Prior Network: Toward Tighter Multi-oriented Text Detection

类别不平衡目标检测

• Solution for Large-Scale Hierarchical Object Detection Datasets with Incomplete Annotation and Data Imbalance

小数据集目标检测

• Comparison Detector: A novel object detection method for small dataset

弱监督目标检测

• Weakly Supervised Deep Detection Networks

非极大值抑制

• Learning non-maximum suppression
• Improving Object Detection With One Line of Code，soft-nms

困难样本采样策略

• Loss Rank Mining: A General Hard Example Mining Method for Real-time Detectors

回归框loss

目前常用的回归框loss有l2 loss，smooth l1 loss和IoU loss。

网络实现

• SSD，ssd_understanding
• Faster RCNN，faster_rcnn_understanding
• R-FCN，rfcn_understanding
• ...

非极大值抑制

• soft-nms

困难样例学习

• Training Region-based Object Detectors with Online Hard Example Mining，非常有效的针对Regin-based目标检测模型的在线困难样例学习策略。

数据集实现

• COCO
• ...

用法

可视化

visdom

# 在tmux或者另一个终端中开启可视化服务器visdom
python -m visdom.server
# 然后在浏览器中查看127.0.0.1:9097

训练

# 训练模型
python train.py

校验

# 校验模型
python validate.py

测试

# 测试模型
python test.py

TODO

• 实现数据集加载VOC