Download as PDF, PPTX
![Average Precision (AP)
● [In the vision community] AP is the estimated area under the PR curve](https://image.slidesharecdn.com/introductiontoobjectdetection-180326063257/75/Introduction-to-object-detection-20-2048.jpg)
Uploaded byBrodmann17
Introduction to object detection
The document outlines the fundamentals of object detection in computer vision, including classification, localization, and evaluation metrics like Intersection over Union (IoU) and Mean Average Precision (mAP). It details the evolution of object detection techniques from 2001 to the present and discusses various datasets used in the field, highlighting the increase in complexity and object instances over time. The document also emphasizes the importance of metrics for evaluating detection algorithms and mentions emerging metrics for further refining evaluation processes.
In this document
Powered by AI
Overview of object detection as an important computer vision task.
Discusses various computer vision tasks with a focus on object detection as introduced in CS231n.
Defines classification, localization, and their combined approach with focus on loss functions and metrics.
Describes ImageNet dataset specifics, including 1000 classes and 800 training images per class.
Explains IoU as a critical measurement for evaluating object localization accuracy.
Details on model architectures comprising classification and localization heads with specific outputs.
Key developments in object detection from 2001 to 2012 highlighting important models and papers.
Explores the state of object detection research and technologies as of the presentation date.
Overview of major datasets like Pascal VOC, ImageNet ILSVRC, and MS COCO with key statistics.
Visual examples from the Pascal VOC and COCO datasets illustrating object detection challenges.
Summarizes the input and output format of object detection algorithms, focusing on predicted detections.
Comprehensive discussion on evaluation metrics including true/false positives, mean average precision, and emerging trends.
More Related Content
![[PR12] You Only Look Once (YOLO): Unified Real-Time Object Detection](https://cdn.slidesharecdn.com/ss_thumbnails/yolo-170616085751-thumbnail.jpg?width=640&height=640&fit=bounds)
Similar to Introduction to object detection
![Lecture 5,6(Comp Vision) [Auto-saved].pptx](https://cdn.slidesharecdn.com/ss_thumbnails/lecture56compvisionauto-saved-250415183559-508e5342-thumbnail.jpg?width=640&height=640&fit=bounds)
Introduction to object detection
- 1. Introduction to ObjectDetection
- 2. Computer Vision Tasks Source:CS231n Object detection http://cs231n.stanford.edu/slides/2016/winter1516_lecture8.pdf
- 3. Computer Vision Tasks Source:CS231n Object detection http://cs231n.stanford.edu/slides/2016/winter1516_lecture8.pdf
- 4. Classification + Localization ●Classification: ○ Input: Image ○ Output: Class label ○ Loss: Cross entropy (Softmaxlog) ○ Evaluation metric: Accuracy ● Localization: ○ Input: Image ○ Output: Box in the image (x, y, w, h) ○ Loss: L2 Loss (Euclidean distance) ○ Evaluation metric: Intersection over Union ● Classification + Localization: ○ Input: Image ○ Output: Class label + box in the image ○ Loss: Sum of both losses
- 5. Classification + Localization:ImageNet Challenge ● Dataset ○ 1000 Classes. ○ Each image has 1 class with at least one bounding box. ○ ~800 Training images per class. ● Evaluation ○ Algorithm produces 5 (class + bounding box) guesses. ○ Example is correct if at least one of guess has correct class AND bounding box at least 50% intersection over union.
- 6. Intersection Over Union(IoU) ● Important measurement for object localization. ● Used in both training and evaluation. Intersection(A,B) Union(A,B) IoU(A,B) =
- 7. Classification + Localization:Model Classification Head: ● C Scores for C classes Localization Head: ● Class agnostic: (x,y,w,h) ● Class specific: (x,y,w,h) X C
- 8. Computer Vision Tasks Source:CS231n Object detection http://cs231n.stanford.edu/slides/2016/winter1516_lecture8.pdf
- 9. Object Detection 2001-2007 ●Rapid Object Detection using a Boosted Cascade of Simple Features (2001) ○ Viola & Jones ● Histograms of Oriented Gradients for Human Detection (2005) ○ Dalal & Triggs ● Object Detection with Discriminatively Trained Part Based Models (2010) ○ Felzenszwalb, Girshick, Ramanan ● Fast Feature Pyramids for Object Detection (2014) ○ Dollar
- 10. Object Detection 2007-2012 Source:Ross Girshick’s CVPR 2017 Tutorial http://deeplearning.csail.mit.edu/instance_ross.pptx
- 11. Object Detection Today Source:Ross Girshick’s CVPR 2017 Tutorial http://deeplearning.csail.mit.edu/instance_ross.pptx
- 12. Object Detection: Datasets 2007 PascalVOC ● 20 Classes ● 11K Training images ● 27K Training objects Was de-facto standard, currently used as quick benchmark to evaluate new detection algorithms. 2013 ImageNet ILSVRC ● 200 Classes ● 476K Training images ● 534K Training objects Essentially scaled up version of PASCAL VOC, similar object statistics. 2015 MS COCO ● 80 Classes ● 200K Training images ● 1.5M Training objects More categories and more object instances in every image. Only 10% of images contain a single object category, 60% in Pascal. More small objects than large objects.
- 13.
- 14.
- 15. Object Detection ● Input:Image ● Output: For each object class c and each image i, an algorithm returns predicted detections: locations with confidence scores .
- 16. Object Detection: Evaluation ●True positive: correct class prediction AND IoU > 50%. ● False positive: wrong class or IoU < 50%. ● False negative: missed (not detected) object ● Only one detection can be matched to an object. ●
- 17. Object Detection: Evaluation ●Mean Average Precision (mAP) across all classes, based on Average Precision (AP) per class, based on Precision and Recall.
- 18. Precision And RecallFor a Threshold
- 19. Precision-Recall Curve Source: Drawingby Prof. William H. Press, the University of Texas at Austin
- 20. Average Precision (AP) ●[In the vision community] AP is the estimated area under the PR curve
- 21. Mean Average Precision(mAP) ● The winner of each object class is the team with the highest average precision ● The winner of the challenge is the team with the highest mean Average Precision (mAP) across all classes.
- 22. Object Detection: Evaluation ●Mean Average Precision (mAP) across all classes, based on Average Precision (AP) per class, based on Precision and Recall.
- 23. Object Detection: Evaluation ●Today new metrics are emerging ○ Averaging precision over all IoU thresholds: 0.5:0.05:0.95 ○ Averaging precision for different object sizes (small, medium, big) ○ Averaging recall as a metric to measure object proposal quality.
- 24. Looking for brilliantresearchers [email protected]