From 651f74ca26290686d7180f60cc0f839c0f784c4e Mon Sep 17 00:00:00 2001 From: Richard Kuo Date: Fri, 2 Aug 2024 05:22:13 +0800 Subject: [PATCH] Add files via upload --- _posts/2023-12-05-Object-Detection.md | 657 ++++++++++++++++++++++++++ 1 file changed, 657 insertions(+) create mode 100644 _posts/2023-12-05-Object-Detection.md diff --git a/_posts/2023-12-05-Object-Detection.md b/_posts/2023-12-05-Object-Detection.md new file mode 100644 index 00000000..c22748b5 --- /dev/null +++ b/_posts/2023-12-05-Object-Detection.md @@ -0,0 +1,657 @@ +--- +layout: post +title: Object Detection +author: [Richard Kuo] +category: [Lecture] +tags: [jekyll, ai] +--- + +Introduction to Image Datasets, Object Detection, Object Tracking, and its Applications. + +--- +## Datasets + +### [COCO Dataset](https://cocodataset.org/) +![](https://cocodataset.org/images/coco-examples.jpg) +* Object segmentation +* Recognition in context +* Superpixel stuff segmentation +* 330K images (>200K labeled) +* 1.5 million object instances +* **80** object categories +* 91 stuff categories +* 5 captions per image +* 250,000 people with keypoints + +--- +### [Open Images Dataset](https://storage.googleapis.com/openimages/web/index.html) +* 15,851,536 boxes on **600** categories +* 2,785,498 instance segmentations on **350** categories +* 3,284,280 relationship annotations on 1,466 relationships +* 66,391,027 point-level annotations on 5,827 classes +* 61,404,966 image-level labels on 20,638 classes +* Extension - 478,000 crowdsourced images with 6,000+ categories + +--- +### [labelme](https://github.com/labelmeai/labelme) +![](https://github.com/labelmeai/labelme/raw/main/examples/instance_segmentation/.readme/annotation.jpg) +`pip install labelme`
+`labelme pic123.jpg`
+ +### [Labelme2YOLO](https://github.com/rooneysh/Labelme2YOLO) +`pip install labelme2yolo`
+ +* Convert JSON files, split training and validation dataset by --val_size
+`python labelme2yolo.py --json_dir /home/username/labelme_json_dir/ --val_size 0.2`
+ +--- +### [LabelImg](https://github.com/tzutalin/labelImg) +![](https://raw.githubusercontent.com/tzutalin/labelImg/master/demo/demo3.jpg) +`pip install labelImg`
+ +`labelImg`
+`labelImg [IMAGE_PATH] [PRE-DEFINED CLASS FILE]`
+ +--- +### VOC .xml convert to YOLO .txt +`cd ~/tf/raccoon/annotations` +`python ~/tf/xml2yolo.py` + +--- +### YOLO Annotation formats (.txt) +**class_num x, y, w, h**
+``` +0 0.5222826086956521 0.5518115942028986 0.025 0.010869565217391304 +0 0.5271739130434783 0.5057971014492754 0.013043478260869565 0.004347826086956522 +``` + +--- +## Object Detection + +![](https://www.researchgate.net/profile/Zhengxia-Zou-2/publication/333077580/figure/fig2/AS:758306230501380@1557805702766/A-road-map-of-object-detection-Milestone-detectors-in-this-figure-VJ-Det-10-11-HOG.ppm) +

+ +--- +### Object Detection Landscape +**Blog:** [The Object Detection Landscape: Accuracy vs Runtime](https://deci.ai/blog/object-detection-landscape-accuracy-vs-runtime/)
+![](https://blog.roboflow.com/content/images/2020/06/image-10.png?raw=true) + +--- +### R-CNN, Fast R-CNN, Faster R-CNN +**Blog:** [目標檢測](https://www.twblogs.net/a/5cb52483bd9eee0f00a1ad24) + +![](https://pic1.xuehuaimg.com/proxy/csdn/https://img-blog.csdnimg.cn/20190415130546284.png) + +* **R-CNN**首先使用Selective search提取region proposals（候選框）；然後用Deep Net（Conv layers）進行特徵提取；最後對候選框類別分別採用SVM進行類別分類，採用迴歸對bounding box進行調整。其中每一步都是獨立的。 +* **Fast R-CNN**在R-CNN的基礎上，提出了多任務損失(Multi-task Loss), 將分類和bounding box迴歸作爲一個整體任務進行學習；另外，通過ROI Projection可以將Selective Search提取出的ROI區域（即：候選框Region Proposals）映射到原始圖像對應的Feature Map上，減少了計算量和存儲量，極大的提高了訓練速度和測試速度。 +* **Faster R-CNN**則是在Fast R-CNN的基礎上，提出了RPN網絡用來生成Region Proposals。通過網絡共享將提取候選框與目標檢測結合成一個整體進行訓練，替換了Fast R-CNN中使用Selective Search進行提取候選框的方法，提高了測試過程的速度。 + +--- +### R-CNN +**Paper:** [arxiv.org/abs/1311.2524](https://arxiv.org/abs/1311.2524)
+![](https://miro.medium.com/max/700/1*REPHY47zAyzgbNKC6zlvBQ.png) +![](https://miro.medium.com/max/500/1*E-8oQW8ZO-hHgTf6laWhhQ.png) + +--- +### Fast R-CNN +**Paper:** [arxiv.org/abs/1504.08083](https://arxiv.org/abs/1504.08083)
+**Github:** [faster-rcnn](https://github.com/rbgirshick/fast-rcnn)
+![](https://miro.medium.com/max/700/1*0pMP3aY8blSpva5tvWbnKA.png) + +--- +### Faster R-CNN +**Paper:** [arxiv.org/abs/1506.01497](https://arxiv.org/abs/1506.01497)
+**Github:** [faster_rcnn](https://github.com/ShaoqingRen/faster_rcnn), [py-faster-rcnn](https://github.com/rbgirshick/py-faster-rcnn) +![](https://miro.medium.com/max/842/1*ndYVI-YCEGCoyRst1ytHjA.png) + +--- +**Blog:** [[物件偵測] S3: Faster R-CNN 簡介](https://ivan-eng-murmur.medium.com/object-detection-s3-faster-rcnn-%E7%B0%A1%E4%BB%8B-5f37b13ccdd2)
+* RPN是一個要提出proposals的小model，而這個小model需要我們先訂出不同尺度、比例的proposal的邊界匡的雛形。而這些雛形就叫做anchor。 + +

+

+ +* RPN的上路是負責判斷anchor之中有無包含物體的機率，因此，1×1的卷積深度就是9種anchor，乘上有無2種情況，得18。而下路則是負責判斷anchor的x, y, w, h與ground truth的偏差量(offsets)，因此9種anchor，乘上4個偏差量(dx, dy, dw, dh)，得卷積深度為36。 + +![](https://miro.medium.com/max/1400/1*Fg7DVdvF449PfX5Fd6oOYA.png) + +--- +### Mask R-CNN +**Paper:** [arxiv.org/abs/1703.06870](https://arxiv.org/abs/1703.06870)
+

+

+

+ +**Blog:** [[物件偵測] S9: Mask R-CNN 簡介](https://ivan-eng-murmur.medium.com/%E7%89%A9%E4%BB%B6%E5%81%B5%E6%B8%AC-s9-mask-r-cnn-%E7%B0%A1%E4%BB%8B-99370c98de28)
+

+

+**Code:** [matterport/Mask_RCNN](https://github.com/matterport/Mask_RCNN)
+

+

+

) + +--- +### SSD: Single Shot MultiBox Detector +**Paper:** [arxiv.org/abs/1512.02325](https://arxiv.org/abs/1512.02325)
+**Blog:** [Understanding SSD MultiBox — Real-Time Object Detection In Deep Learning](https://towardsdatascience.com/understanding-ssd-multibox-real-time-object-detection-in-deep-learning-495ef744fab)
+![](https://miro.medium.com/max/700/1*JuhjYUWXgfxMMoa4SIKLkA.png) +使用神經網絡（VGG-16）提取feature map後進行分類和回歸來檢測目標物體。 +![](https://miro.medium.com/max/700/1*51joMGlhxvftTxGtA4lA7Q.png) +![](https://miro.medium.com/max/480/1*IZf0wajQ75DPsoBkWjwlsA.gif) +**Code:** [pierluigiferrari/ssd_keras](https://github.com/pierluigiferrari/ssd_keras)
+ + + + + + + + + +
+ +--- +### RetinaNet +**Paper:** [Focal Loss for Dense Object Detection](https://arxiv.org/abs/1708.02002)
+**Blog:** [RetinaNet 介紹](https://gino6178.medium.com/%E7%89%A9%E4%BB%B6%E5%81%B5%E6%B8%AC-retinanet-%E4%BB%8B%E7%B4%B9-dda4100673bb) +![](https://miro.medium.com/max/622/0*ksQqcCYF0iQN_oX2.png) +從左到右分別用上了
+* 殘差網路(Residual Network ResNet) +* 特徵金字塔(Feature Pyramid Network FPN) +* 類別子網路(Class Subnet) +* 框子網路(Box Subnet) +* 以及Anchors + +**Blog:** [Review: RetinaNet — Focal Loss](https://towardsdatascience.com/review-retinanet-focal-loss-object-detection-38fba6afabe4) +![](https://miro.medium.com/max/600/0*E30eIZ5aCGjhCz9E.gif) + +**Code:** [keras-retinanet](https://github.com/fizyr/keras-retinanet)
+ +--- +### CornerNet +**Paper:** [CornerNet: Detecting Objects as Paired Keypoints](https://arxiv.org/abs/1808.01244)
+![](https://timy90022.github.io/2019/08/09/CornerNet-Detecting-Objects-as-Paired-Keypoints/1.png) +**Code:** [princeton-vl/CornerNet](https://github.com/princeton-vl/CornerNet)
+ +--- +### CenterNet +**Paper:** [CenterNet: Keypoint Triplets for Object Detection](https://arxiv.org/abs/1904.08189)
+![](https://deci.ai/wp-content/uploads/2021/05/7-Architecture-of-CenterNet.png) +**Code:** [xingyizhou/CenterNet](https://github.com/xingyizhou/CenterNet)
+ +--- +### EfficientDet +**Paper:** [arxiv.org/abs/1911.09070](https://arxiv.org/abs/1911.09070)
+![](https://github.com/google/automl/raw/master/efficientdet/g3doc/network.png) +**Code:** [google efficientdet](https://github.com/google/automl/tree/master/efficientdet)
+

+ + + + + +
+**Kaggle:** [rkuo2000/efficientdet-gwd](https://www.kaggle.com/rkuo2000/efficientdet-gwd)
+![](https://camo.githubusercontent.com/f0c80711512aacc0c1423a64e1036608a773f61c0bb6480ec0d57614ce3e7ccc/68747470733a2f2f696d6775722e636f6d2f3579554a4350562e6a7067) + +--- +### [YOLO- You Only Look Once](https://pjreddie.com/darknet/yolo/) +**Code:** [pjreddie/darknet](https://github.com/pjreddie/darknet)
+![](https://pyimagesearch.com/wp-content/uploads/2018/11/yolo_design.jpg) + +**[YOLOv1](https://arxiv.org/abs/1506.02640)** : mapping bounding box
+![](https://manalelaidouni.github.io/assets/img/pexels/YOLO_arch.png) + +**[YOLOv2](https://arxiv.org/abs/1612.08242)** : anchor box proportional to K-means
+![](https://2.bp.blogspot.com/-_R-w_tWHdzc/WzJPsol7qFI/AAAAAAABbgg/Jsf-AO3qH0A9oiCeU0LQxN-wdirlOz4WgCLcBGAs/s400/%25E8%259E%25A2%25E5%25B9%2595%25E5%25BF%25AB%25E7%2585%25A7%2B2018-06-26%2B%25E4%25B8%258B%25E5%258D%258810.36.51.png) + +**[YOLOv3](https://arxiv.org/abs/1804.02767)** : Darknet-53 + FPN
+![](https://media.springernature.com/m685/springer-static/image/art%3A10.1038%2Fs41598-021-81216-5/MediaObjects/41598_2021_81216_Fig1_HTML.png) +![](https://miro.medium.com/max/2000/1*d4Eg17IVJ0L41e7CTWLLSg.png) + +--- +### YOLObile +**Paper:** [arxiv.org/abs/2009.05697](https://arxiv.org/abs/2009.05697)
+**Blog:** [YOLObile：移動設備上的實時目標檢測](https://twgreatdaily.com/zh-hk/jRnp3HQBd8y1i3sJ_wK9.html)
+![](https://images.twgreatdaily.com/images/elastic/vhk/vhk33XQBd8y1i3sJeRJg.jpg) +**Code:** [nightsnack/YOLObile](https://github.com/nightsnack/YOLObile)
+

+ +--- +### YOLOv4 +**Paper:** [YOLOv4: Optimal Speed and Accuracy of Object Detection](https://arxiv.org/abs/2004.10934)
+* YOLOv4 = YOLOv3 + CSPDarknet53 + SPP + PAN + BoF + BoS
+![](https://www.researchgate.net/profile/Jishu-Miao/publication/349381918/figure/fig4/AS:994706955722753@1614168027527/Normal-YOLOv4-network-architecture.ppm) +* CSP +![](https://blog.roboflow.com/content/images/2020/06/image-15.png) +* PANet +![](https://blog.roboflow.com/content/images/2020/06/image-17.png) + +**Code:** [AlexeyAB/darknet](https://github.com/AlexeyAB/darknet)
+**Code:** [WongKinYiu/PyTorch_YOLOv4](https://github.com/WongKinYiu/PyTorch_YOLOv4)
+ +--- +### [YOLOv5](https://docs.ultralytics.com/yolov5/) +

+**Code:** [ultralytics/yolov5/](https://github.com/ultralytics/yolov5/)
+

< img src="https://user-images.githubusercontent.com/26833433/127574988-6a558aa1-d268-44b9-bf6b-62d4c605cc72.jpg">

+

< img src="https://user-images.githubusercontent.com/26833433/136901921-abcfcd9d-f978-4942-9b97-0e3f202907df.png">

+ +--- +### Scaled-YOLOv4 +**Paper:** [arxiv.org/abs/2011.08036](https://arxiv.org/abs/2011.08036)
+![](https://miro.medium.com/max/1838/1*OE4SO1U87DHcAClSZFGlMg.png) +**Code:** [WongKinYiu/ScaledYOLOv4](https://github.com/WongKinYiu/ScaledYOLOv4)
+ +--- +### YOLOR : You Only Learn One Representation +**Paper:** [arxiv.org/abs/2105.04206](https://arxiv.org/abs/2105.04206)
+![](https://github.com/WongKinYiu/yolor/raw/main/figure/unifued_network.png?raw=true) +**Code:** [WongKinYiu/yolor](https://github.com/WongKinYiu/yolor)
+

+![](https://github.com/WongKinYiu/yolor/raw/main/figure/performance.png?raw=true) + +--- +### YOLOX +**Paper:** [arxiv.org/abs/2107.08430](https://arxiv.org/abs/2107.08430)
+![](https://miro.medium.com/max/915/1*ihnRFgPMgatEtrlTtOM2Bg.png) +**Code:** [Megvii-BaseDetection/YOLOX](https://github.com/Megvii-BaseDetection/YOLOX)
+![](https://github.com/Megvii-BaseDetection/YOLOX/blob/main/assets/demo.png?raw=true) +![](https://github.com/Megvii-BaseDetection/YOLOX/blob/main/assets/git_fig.png?raw=true) + +--- +### YOLOv5 vs YOLOX +**Paper:** [Evaluation of YOLO Models with Sliced Inference for Small Object Detection](https://arxiv.org/abs/2203.04799)
+ +

+ +--- +### CSL-YOLO +**Paper:** [arxiv.org/abs/2107.04829](https://arxiv.org/abs/2107.04829)
+![](https://www.researchgate.net/publication/353208773/figure/fig1/AS:1044955464224768@1626148205216/Overall-architecture-of-CSL-YOLO-the-convolution-1x1-is-weights-sharing.ppm) +**Code:** [D0352276/CSL-YOLO](https://github.com/D0352276/CSL-YOLO)
+![](https://github.com/D0352276/CSL-YOLO/blob/main/demo/result_img_1.png?raw=true) +**Camera Demo**
+![](https://github.com/D0352276/CSL-YOLO/blob/main/demo/camera_demo.gif?raw=true) + +--- +### PP-YOLOE +**Paper:** [PP-YOLOE: An evolved version of YOLO](https://arxiv.org/abs/2203.16250)
+![](https://github.com/rkuo2000/AI-course/blob/main/images/PP-YOLOE.png?raw=true) +**Code:** [PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection)
+![](https://github.com/rkuo2000/AI-course/blob/main/images/PP-YOLOE_MS_COCO.png?raw=true) +**Kaggle:** [rkuo2000/pp-yoloe](https://www.kaggle.com/code/rkuo2000/pp-yoloe)
+![](https://github.com/rkuo2000/AI-course/blob/main/images/PP-YOLOE_demo.jpg?raw=true) + +--- +### YOLOv6 +**Blog:** [YOLOv6：又快又准的目标检测框架开源啦](https://tech.meituan.com/2022/06/23/yolov6-a-fast-and-accurate-target-detection-framework-is-opening-source.html)
+* RegVGG是一種簡單又强力的CNN結構，在訓練時使用了性能高的多分支模型，而在推理時使用了速度快、省内存的單路模型，也是更具備速度和精度的均衡。 +

+ +* EfficientRep將在backbone中stride=2的卷積層换成了stride=2的RepConv層。並且也將CSP-Block修改為RepBlock +![](https://p0.meituan.net/travelcube/8ec8337d37c2545b8fcf355625854802145939.png) +* 同樣為了降低在硬體上的延遲，在Neck上的特徵融合結構中也引入了Rep結構。在Neck中使用的是Rep-PAN。 +![](https://p0.meituan.net/travelcube/c37c23c37fd094e05e8cab924659a9d9199592.png) +* 和YOLOX一樣，YOLOv6也對檢測頭近行了解耦，分開了邊框回歸與類别分類的過程。 +![](https://pic4.zhimg.com/80/v2-3e5868d7f1aadc76d9b112e4cb79719b_720w.jpg) + +**Code:** [meituan/YOLOv6](https://github.com/meituan/YOLOv6)
+![](https://github.com/meituan/YOLOv6/raw/main/assets/picture.png) +![](https://p0.meituan.net/travelcube/bc0e60516ae0bcad1c111d7c0c5c3b9e335568.png) + + +--- +### YOLOv7 +**Paper:** [YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors](https://arxiv.org/abs/2207.02696)
+* Extended efficient layer aggregation networks +![](https://github.com/rkuo2000/AI-course/blob/main/images/YOLOv7_E-ELAN.png?raw=true) +* Model scaling for concatenation-based models +![](https://github.com/rkuo2000/AI-course/blob/main/images/YOLOv7_model_scaling.png?raw=true) +* Planned re-parameterized convolution +![](https://github.com/rkuo2000/AI-course/blob/main/images/YOLOv7_planned_reparameterized_model.png?raw=true) +* Coarse for auxiliary and fine for lead head label assigner +![](https://github.com/rkuo2000/AI-course/blob/main/images/YOLOv7_coarse_to_fine_lead_guided_assigner.png?raw=true) + +**Code:** [WongKinYiu/yolov7](https://github.com/WongKinYiu/yolov7)
+

+ +--- +### YOLOv8 +[Ultralytics YOLOv8](https://www.ultralytics.com/) is a cutting-edge, state-of-the-art (SOTA) model that builds upon the success of previous YOLO versions and introduces new features and improvements to further boost performance and flexibility. YOLOv8 is designed to be fast, accurate, and easy to use, making it an excellent choice for a wide range of object detection and tracking, instance segmentation, image classification and pose estimation tasks. + +**Blog:** [Dive into YOLOv8](https://openmmlab.medium.com/dive-into-yolov8-how-does-this-state-of-the-art-model-work-10f18f74bab1)
+**Paper:** [Real-Time Flying Object Detection with YOLOv8](https://arxiv.org/abs/2305.09972)
+![](https://user-images.githubusercontent.com/27466624/239739723-57391d0f-1848-4388-9f30-88c2fb79233f.jpg) + +**Code:** [https://github.com/ultralytics/ultralytics](https://github.com/ultralytics/ultralytics)
+**Kaggle:**
+* [https://www.kaggle.com/code/rkuo2000/yolov8](https://www.kaggle.com/code/rkuo2000/yolov8) +* [https://www.kaggle.com/code/rkuo2000/yolov8-pothole-detection](https://www.kaggle.com/code/rkuo2000/yolov8-pothole-detection) + +--- +### UAV-YOLOv8 +**Paper:** [UAV-YOLOv8: A Small-Object-Detection Model Based on Improved YOLOv8 for UAV Aerial Photography Scenarios](https://www.mdpi.com/1424-8220/23/16/7190)
+ +--- +### YOLOv8 Aerial Sheep Detection and Counting +**Code:** [https://github.com/monemati/YOLOv8-Sheep-Detection-Counting](https://github.com/monemati/YOLOv8-Sheep-Detection-Counting)
+

+ +--- +### YOLOv8 Drone Surveillance +**Code:** [https://github.com/ni9/Object-Detection-From-Drone-For-Surveillance](https://github.com/ni9/Object-Detection-From-Drone-For-Surveillance)
+![](https://github.com/ni9/Object-Detection-From-Drone-For-Surveillance/raw/main/images/Capture1.PNG) + +--- +### YOLOv9 +**Paper:** [YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information](https://arxiv.org/abs/2402.13616)
+ +**Blog:** [YOLOv9: Advancing the YOLO Legacy](https://learnopencv.com/yolov9-advancing-the-yolo-legacy/)
+![](https://learnopencv.com/wp-content/uploads/2024/02/feature-1.gif) +**Programmable Gradient Information (PGI)** +![](https://learnopencv.com/wp-content/uploads/2024/02/pgi1-1024x434.png) +**GELAN architecture**
+![](https://learnopencv.com/wp-content/uploads/2024/02/glean-1024x432.png) + +**Code:** [https://github.com/WongKinYiu/yolov9](https://github.com/WongKinYiu/yolov9)
+![](https://github.com/WongKinYiu/yolov9/raw/main/figure/performance.png) + +--- +### YOLOv10 +**Paper:** [YOLOv10: Real-Time End-to-End Object Detection](https://arxiv.org/abs/2405.14458)
+**Code:** [https://github.com/THU-MIG/yolov10](https://github.com/THU-MIG/yolov10)
+![](https://github.com/ultralytics/ultralytics/assets/26833433/e0360eb4-3589-4cd1-b362-a8970bceada6) + +--- +## Trash Detection +### Localize and Classify Wastes on the Streets +**Paper:** [arxiv.org/abs/1710.11374](https://arxiv.org/abs/1710.11374)
+**Model:** GoogLeNet
+![](https://d3i71xaburhd42.cloudfront.net/5e409a99833470206dac6cf79a4f857d5436dd4a/2-Figure1-1.png) + +--- +### Street Litter Detection +**Code:** [isaychris/litter-detection-tensorflow](https://github.com/isaychris/litter-detection-tensorflow)
+![](https://camo.githubusercontent.com/ab04d9b6af8e7885d44eb001f38c82a9682c8132a44648f6224eaa393cfba080/68747470733a2f2f692e696d6775722e636f6d2f456671716f536d2e706e67) + +--- +### [TACO: Trash Annotations in Context](http://tacodataset.org/) +**Paper:** [arxiv.org/abs/2003.06875](https://arxiv.org/abs/2003.06875)
+**Code:** [pedropro/TACO](https://github.com/pedropro/TACO)
+**Model:** Mask R-CNN +![](https://raw.githubusercontent.com/wiki/pedropro/TACO/images/teaser.gif) + +--- +### Marine Litter Detection +**Paper:** [arxiv.org/abs/1804.01079](https://arxiv.org/abs/1804.01079)
+**Dataset:** [Deep-sea Debris Database](http://www.godac.jamstec.go.jp/catalog/dsdebris/e/)
+![](https://d3i71xaburhd42.cloudfront.net/aa9ca01584600207773814660d8ba20a8a830772/6-Figure3-1.png) + +--- +### Marine Debris Detection +**Ref.** [Detect Marine Debris from Aerial Imagery](https://medium.com/@yhoso/mapping-marine-debris-with-keras-part-1-f485dedf2073)
+**Code:** [yhoztak/object_detection](https://github.com/yhoztak/object_detection)
+**Model:** RetinaNet +![](https://miro.medium.com/max/700/1*EtGCA8Bux9xJcUaHgs63IA.png) +![](https://miro.medium.com/max/700/1*8hi2MeOFBCNA4B_33I7VaA.png) + +--- +### UDD dataset +**Paper:** [A New Dataset, Poisson GAN and AquaNet for Underwater Object Grabbing](https://arxiv.org/abs/2003.01446)
+**Dataset:** [UDD_Official](https://github.com/chongweiliu/UDD_Official)
+Concretely, UDD consists of 3 categories (seacucumber, seaurchin, and scallop) with 2,227 images +![](https://github.com/chongweiliu/UDD_Official/raw/main/results.jpg?raw=true) +![](https://d3i71xaburhd42.cloudfront.net/7edd63a0668014c825a702a156e8aea4e527d57a/2-Figure2-1.png) +![](https://d3i71xaburhd42.cloudfront.net/7edd63a0668014c825a702a156e8aea4e527d57a/4-Figure4-1.png) + +--- +### Detecting Underwater Objects (DUO) +**Paper:** [A Dataset And Benchmark Of Underwater Object Detection For Robot Picking](https://arxiv.org/abs/2106.05681)
+**Dataset:** [DUO](https://drive.google.com/file/d/1w-bWevH7jFs7A1bIBlAOvXOxe2OFSHHs/view)
+![](https://d3i71xaburhd42.cloudfront.net/5951ed58d17cc510dd32da3db47c4f0fed08b80e/2-Figure1-1.png) + +--- +## Object Tracking Datasets +**Paper:** [Deep Learning in Video Multi-Object Tracking: A Survey](https://arxiv.org/abs/1907.12740)
+ +### [Multiple Object Tracking (MOT)](https://motchallenge.net/) +**[MOT-16](https://motchallenge.net/data/MOT16/)**
+![](https://d3i71xaburhd42.cloudfront.net/ac0d88ca5f75a4a80da90365c28fa26f1a26d4c4/3-Figure1-1.png) + +--- +### Under-water Ojbect Tracking (UOT) +**Paper:** [Underwater Object Tracking Benchmark and Dataset](http://www.hstabstractbook.org/index_htm_files/c-PID6132325.pdf)
+**[UOT32](https://www.kaggle.com/landrykezebou/uot32-underwater-object-tracking-dataset)**
+**[UOT100](https://www.kaggle.com/landrykezebou/uot100-underwater-object-tracking-dataset)**
+![](https://github.com/rkuo2000/AI-course/blob/main/images/UOT32.png?raw=true) + +--- +### Re3 : Real-Time Recurrent Regression Networks for Visual Tracking of Generic Objects +**Paper:** [arxiv.org/abs/1705.06368](https://arxiv.org/abs/1705-06368)
+**Code:** [moorejee/Re3](https://github.com/moorejee/Re3)
+![](https://github.com/moorejee/Re3/blob/master/demo/output.gif?raw=true) + +--- +### Deep SORT +**Paper:** [arxiv.org/abs/1703.07402](https://arxiv.org/abs/1703.07402)
+**Code:** [nwojke/deep_sort](https://github.com/nwojke/deep_sort)
+**Blog:** [Deep SORT多目标跟踪算法代码解析(上)](https://zhuanlan.zhihu.com/p/133678626)
+* **Kalman Filter** to create “Track”, associate track_i with incoming detection_k +* A distance metric (**squared Mahalanobis distance**) to quantify the association +* an efficient algorithm (**standard Hungarian algorithm**) to associate the data + +--- +### SiamCAR +**Paper:** [arxiv.org/abs/1911.07241](https://arxiv.org/abs/1911.07241)
+**Code:** [ohhhyeahhh/SiamCAR](https://github.com/ohhhyeahhh/SiamCAR)
+![](https://media.arxiv-vanity.com/render-output/5247410/x2.png) +![](https://media.arxiv-vanity.com/render-output/5247410/x1.png) + +--- +### YOLOv5 + DeepSort +**Code:** [HowieMa/DeepSORT_YOLOv5_Pytorch](https://github.com/HowieMa/DeepSORT_YOLOv5_Pytorch)
+![](https://github.com/mikel-brostrom/Yolov5_StrongSORT_OSNet/raw/master/MOT16_eval/track_pedestrians.gif?raw=true) + +--- +### Yolov5 + StrongSORT with OSNet +**Code:** [Yolov5_StrongSORT_OSNet](https://github.com/mikel-brostrom/Yolov5_StrongSORT_OSNet)
+ + + + + +
+ +--- +### SiamBAN +**Paper:** [arxiv.org/abs/2003.06761](https://arxiv.org/abs/2003.06761)
+**Code:** [hqucv/siamban](https://github.com/hqucv/siamban)
+**Blog:** [[CVPR2020][SiamBAN] Siamese Box Adaptive Network for Visual Tracking](https://www.bilibili.com/read/cv7541809) +![](https://i0.hdslb.com/bfs/article/357345f94693ef09cd71406530f42c590a756336.png@942w_444h_progressive.webp) +![](https://github.com/hqucv/siamban/blob/master/demo/output/12.gif?raw=true) +![](https://github.com/hqucv/siamban/blob/master/demo/output/34.gif?raw=true) + +--- +### FairMOT +**Paper:** [FairMOT: On the Fairness of Detection and Re-Identification in Multiple Object Tracking](https://arxiv.org/abs/2004.01888)
+**Code:** [ifzhang/FairMOT](https://github.com/ifzhang/FairMOT)
+![](https://github.com/ifzhang/FairMOT/blob/master/assets/pipeline.png?raw=true) + +--- +### 3D-ZeF +**Paper:** [arxiv.org/abs/2006.08466](https://arxiv.org/abs/2006.08466)
+**Code:** [mapeAAU/3D-ZeF](https://github.com/mapeAAU/3D-ZeF)
+![](https://vap.aau.dk/wp-content/uploads/2020/05/setup-300x182.png) +![](https://vap.aau.dk/wp-content/uploads/2020/05/bitmap.png) + +--- +## UAV-based Object Detection and Tracking +**Paper:** [Deep Learning for UAV-based Object Detection and Tracking: A Survey](https://arxiv.org/abs/2110.12638)
+ +--- +### Efficient Object Detection Model for Real-Time UAV Applications +**Paper:** [Efficient Object Detection Model for Real-Time UAV Applications](https://arxiv.org/abs/1906.00786)
+ +--- +## Applications +[Satellite Image Deep Learning](https://github.com/robmarkcole/satellite-image-deep-learning)
+ +### T-CNN : Tubelets with CNN +**Paper:** [arxiv.org/abs/1604.02532](https://arxiv.org/abs/1604.02532)
+**Blog:** [人工智慧在太空的應用](https://www.narlabs.org.tw/xcscience/cont?xsmsid=0I148638629329404252&qcat=0I164512713411182211&sid=0J295566068384018349)
+![](https://www.narlabs.org.tw/files/file_pool/1/0J295568342633298375/%E5%9C%962.png) +![](https://www.narlabs.org.tw/files/file_pool/1/0J295570208893834419/%E5%9C%963.png) +![](https://www.narlabs.org.tw/files/file_pool/1/0J295570971338287463/%E5%9C%965.png) + +--- +### Swimming Pool Detection +**Dataset:** [Aerial images of swimming pools](https://www.kaggle.com/datasets/alexj21/swimming-pool-512x512)
+**Kaggle:** [Evaluation Efficientdet - Swimming Pool Detection](https://www.kaggle.com/code/alexj21/evaluation-efficientdet-swimming-pool-detection)
+![](https://cdn-images-1.medium.com/max/1200/0*kbWFs2dOILw7abke.png) + +--- +### Identify Military Vehicles in Satellite Imagery +**Blog:** [Identify Military Vehicles in Satellite Imagery with TensorFlow](https://python.plainenglish.io/identifying-military-vehicles-in-satellite-imagery-with-tensorflow-96015634129d)
+**Dataset:** [Moving and Stationary Target Acquisition and Recognition (MSTAR) Dataset](https://www.sdms.afrl.af.mil/index.php?collection=mstar)
+![](https://github.com/NateDiR/sar_target_recognition_deep_learning/raw/main/images/mstar_example.png) +**Code:** [Target Recognition in Sythentic Aperture Radar Imagery Using Deep Learning](https://github.com/NateDiR/sar_target_recognition_deep_learning)
+[script.ipynb](https://github.com/NateDiR/sar_target_recognition_deep_learning/blob/main/script.ipynb)
+ +### [YOLOv5 Detect](https://kaggle.com/rkuo2000/yolov5-detect) +detect image / video + + +--- +### [YOLOv5 Elephant](https://kaggle.com/rkuo2000/yolov5-elephant) +train YOLOv5 for detecting elephant (dataset from OpenImage V6) + + + + + +
+ +--- +### [BCCD Dataset](https://github.com/Shenggan/BCCD_Dataset) +3 classes: RBC (Red Blood Cell), WBC (White Blood Cell), Platelets (血小板)
+**Kaggle:** [https://www.kaggle.com/datasets/surajiiitm/bccd-dataset](https://www.kaggle.com/datasets/surajiiitm/bccd-dataset)
+![](https://github.com/Shenggan/BCCD_Dataset/raw/master/example.jpg) + +--- +### [Face Mask Dataset](https://www.kaggle.com/datasets/ashishjangra27/face-mask-12k-images-dataset) +![](https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcQk1FR3RwIHt2vX1obpNkJArsK1BXlO96BxJHnuHyo6&s) +**Kaggle:** [https://kaggle.com/rkuo2000/yolov5-facemask](https://kaggle.com/rkuo2000/yolov5-facemask)
+![](https://github.com/rkuo2000/AI-course/blob/main/images/YOLOv5_Facemask.jpg?raw=true) + +--- +### Traffic Analysis +**Kaggle:** [https://kaggle.com/rkuo2000/yolov5-traffic-analysis](https://kaggle.com/rkuo2000/yolov5-traffic-analysis)
+![](https://github.com/rkuo2000/AI-course/blob/main/images/YOLOv5_traffic_analysis.jpg?raw=true) + +--- +### Global Wheat Detection +**Kaggle:** [https://www.kaggle.com/rkuo2000/yolov5-global-wheat-detection](https://www.kaggle.com/rkuo2000/yolov5-global-wheat-detection) +![](https://github.com/rkuo2000/AI-course/blob/main/images/YOLOv5_GWD.jpg?raw=true) +**Kaggle:** [https://www.kaggle.com/rkuo2000/efficientdet-gwd](https://www.kaggle.com/rkuo2000/efficientdet-gwd)
+![](https://github.com/rkuo2000/AI-course/blob/main/images/EfficientDet_GWD.png?raw=true) + +--- +### Mask R-CNN +**Kaggle:** [rkuo2000/mask-rcnn](https://www.kaggle.com/rkuo2000/mask-rcnn) +![](https://github.com/rkuo2000/AI-course/blob/main/images/Mask_RCNN_TF2.png?raw=true) + +--- +### Mask R-CNN transfer learning +**Kaggle:** [Mask RCNN transfer learning](https://www.kaggle.com/hmendonca/mask-rcnn-and-coco-transfer-learning-lb-0-155) +![](https://github.com/rkuo2000/AI-course/blob/main/images/Mask_RCNN_transfer_learning.png?raw=true) + +--- +### YOLOv5 + DeepSort +**Kaggle:** [YOLOv5 DeepSort](https://kaggle.com/rkuo2000/yolov5-deepsort)
+ + + +--- +### Objectron +**Kaggle:** [rkuo2000/mediapipe-objectron](https://www.kaggle.com/rkuo2000/mediapipe-objectron)
+![](https://github.com/rkuo2000/AI-course/blob/main/images/Objectron_shoes.png?raw=true) + +--- +### OpenCV-Python play GTA5 +**Ref.** [Reading game frames in Python with OpenCV - Python Plays GTA V](https://pythonprogramming.net/game-frames-open-cv-python-plays-gta-v/)
+**Code:** [Sentdex/pygta5](https://github.com/Sentdex/pygta5)
+ + +--- +### Steel Defect Detection +**Dataset:** [Severstal: Steel Defect Detection](https://www.kaggle.com/c/severstal-steel-defect-detection)
+![](https://diyago.github.io/images/kaggle-severstal/input_data.png) +**Kaggle:** [https://www.kaggle.com/code/jaysmit/u-net (Keras UNet)](https://www.kaggle.com/code/jaysmit/u-net)
+ +--- +### PCB Defect Detection +**Dataset:** [HRIPCB dataset (dropbox)](https://www.dropbox.com/s/h0f39nyotddibsb/VOC_PCB.zip?dl=0)
+![](https://media.springernature.com/lw685/springer-static/image/art%3A10.1038%2Fs41598-022-16302-3/MediaObjects/41598_2022_16302_Fig4_HTML.png?as=webp) + +--- +### Pothole Detection +**Blog:** [Pothole Detection using YOLOv4](https://learnopencv.com/pothole-detection-using-yolov4-and-darknet/?ck_subscriber_id=638701084)
+**Code:** [yolov4_pothole_detection.ipynb](https://github.com/spmallick/learnopencv/blob/master/Pothole-Detection-using-YOLOv4-and-Darknet/jupyter_notebook/yolov4_pothole_detection.ipynb)
+**Kaggle:** [YOLOv7 Pothole Detection](https://www.kaggle.com/code/rkuo2000/yolov7-pothole-detection) +![](https://learnopencv.com/wp-content/uploads/2022/07/Pothole-Detection-using-YOLOv4-and-Darknet.gif) + +--- +### Car Breaking Detection +**Code**: [YOLOv7 Braking Detection](https://github.com/ArmaanSinghSandhu/YOLOv7-Braking-Detection)
+![](https://github.com/ArmaanSinghSandhu/YOLOv7-Braking-Detection/raw/main/results/Detection.gif) + +--- +## Steel Defect Detection +**Dataset:** [Severstal: Steel Defect Detection](https://www.kaggle.com/c/severstal-steel-defect-detection)
+![](https://diyago.github.io/images/kaggle-severstal/input_data.png) + +--- +### Steel Defect Detection using UNet +**Kaggle:** [https://www.kaggle.com/code/jaysmit/u-net (Keras UNet)](https://www.kaggle.com/code/jaysmit/u-net)
+**Kaggle:** [https://www.kaggle.com/code/myominhtet/steel-defection (pytorch UNet](https://www.kaggle.com/code/myominhtet/steel-defection)
+ +--- +### Steel-Defect Detection Using CNN +**Code:** [https://github.com/himasha0421/Steel-Defect-Detection](https://github.com/himasha0421/Steel-Defect-Detection)
+![](https://github.com/himasha0421/Steel-Defect-Detection/raw/main/images/unet_img.png) +![](https://github.com/himasha0421/Steel-Defect-Detection/raw/main/images/img_1.png) + +--- +### MSFT-YOLO +**Paper:** [MSFT-YOLO: Improved YOLOv5 Based on Transformer for Detecting Defects of Steel Surface](https://www.mdpi.com/1424-8220/22/9/3467)
+![](https://www.mdpi.com/sensors/sensors-22-03467/article_deploy/html/images/sensors-22-03467-g002-550.jpg) +![](https://www.mdpi.com/sensors/sensors-22-03467/article_deploy/html/images/sensors-22-03467-g006-550.jpg) + +--- +### PCB Datasets +* [DeepPCB](https://github.com/tangsanli5201/DeepPCB)
+![](https://github.com/tangsanli5201/DeepPCB/raw/master/fig/test.jpg) +* [HRIPCB](https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/joe.2019.1183)
+![](https://ietresearch.onlinelibrary.wiley.com/cms/asset/3130c389-d389-49c9-9ce8-61e624c5a794/tje2bf02982-fig-0003-m.png) + +--- +### PCB Defect Detection +**Paper:** [PCB Defect Detection Using Denoising Convolutional Autoencoders](https://arxiv.org/abs/2008.12589)
+![](https://www.researchgate.net/publication/343986672/figure/fig1/AS:930433990672384@1598844158824/The-overview-of-the-system-a-salt-and-pepper-noise-is-added-to-defective-PCBs-b.ppm) +![](https://www.researchgate.net/publication/343986672/figure/fig4/AS:930433994878976@1598844159541/examples-of-the-output-of-the-proposed-model.ppm) + +--- +### PCB Defect Classification +**Dataset:** [HRIPCB dataset (dropbox)](https://www.dropbox.com/s/h0f39nyotddibsb/VOC_PCB.zip?dl=0)
+印刷电路板（PCB）瑕疵数据集。它是一个公共合成PCB数据集，包含1386张图像，具有6种缺陷（漏孔、鼠咬、开路、短路、杂散、杂铜），用于图像检测、分类和配准任务。
+**Paper:** [End-to-end deep learning framework for printed circuit board manufacturing defect classification](https://www.nature.com/articles/s41598-022-16302-3)
+![](https://media.springernature.com/full/springer-static/image/art%3A10.1038%2Fs41598-022-16302-3/MediaObjects/41598_2022_16302_Fig2_HTML.png?as=webp) +![](https://media.springernature.com/lw685/springer-static/image/art%3A10.1038%2Fs41598-022-16302-3/MediaObjects/41598_2022_16302_Fig4_HTML.png?as=webp) + +
+
+ +*This site was last updated {{ site.time | date: "%B %d, %Y" }}.* + +