Deep Learning based object Detection with YOLO v2

I will briefly go through the the process of YOLOv2

Transcript

1. Jumabek Alikhanov @Information Security Research Lab, Inha University YOLO9000: Better, Faster, Stronger (CVPR 2017, Best Paper Honorable Mention) 1
2. 1. Introduction & Previous Work 2. Better detection performance 3. Faster processing speed 4. Detecting more classes(object types) 5. Conclusion CONTENTS 2
3. Task & Evaluation Metric mAP- mean Avarage Precision 3 https://github.com/rafaelpadilla/Object-Detection-Metrics
4. YOLO v1 Network Output shape = (S, S, B×5 + C) = (7, 7, 2×5 + 20) = (7, 7, 30). 4
5. YOLOv1: Loss Function pi-conditional class Prob. Ci - box confidence score 5 Localization Confidence Classification
6. Previously Pascal 2007 mAP Speed DPM v5 33.7 .07 FPS 14 s/img R-CNN 66.0 .05 FPS 20 s/img Fast R-CNN 70.0 .5 FPS 2 s/img Faster R-CNN 73.2 7 FPS 140 ms/img YOLO 63.4 45 FPS 22 ms/img 6
7. Previously Pascal 2007 mAP Speed DPM v5 33.7 .07 FPS 14 s/img R-CNN 66.0 .05 FPS 20 s/img Fast R-CNN 70.0 .5 FPS 2 s/img Faster R-CNN 73.2 7 FPS 140 ms/img YOLO 63.4 45 FPS 22 ms/img 7
8. Better Performance 8
9. 9 YOLO Train on ImageNet Fine-tune on detection Resize network
10. 10 Fine-tune 448x448 Classifier: +3.5% mAP Train on ImageNet Fine-tune on detection Resize, fine-tune on ImageNet
11. Anchor boxes use static initialization
12. Use k-means clustering to find better initializations https://github.com/Jumabek/darknet_scripts
14. Static Anchors vs Dimension Clusters 14
15. Box Location Prediction 15
16. Dimension Clusters: +5% mAP
17. 17 Multi-scale training: +1.5% mAP
18. YOLOv2: Fast, Accurate Detection
19. Huang, Jonathan, et al. "Speed/accuracy trade-offs for modern convolutional object detectors." arXiv preprint arXiv:1611.10012 (2016).
20. Huang, Jonathan, et al. "Speed/accuracy trade-offs for modern convolutional object detectors." arXiv preprint arXiv:1611.10012 (2016).
21. Huang, Jonathan, et al. "Speed/accuracy trade-offs for modern convolutional object detectors." arXiv preprint arXiv:1611.10012 (2016). YOLOv2
23. Faster Detection Speed 23
24. Speed is not just parameter counts or FLOPs Top 1 Top 5 FLOPs GPU Speed VGG-16 70.5 90.0 30.95 Bn 100 FPS Extraction (YOLOv1) 72.5 90.8 8.52 Bn 180 FPS Resnet50 75.3 92.2 7.66 Bn 90 FPS
25. Darknet19: A good balance of speed and accuracy Top 1 Top 5 FLOPs GPU Speed VGG-16 70.5 90.0 30.95 Bn 100 FPS Extraction (YOLOv1) 72.5 90.8 8.52 Bn 180 FPS Resnet50 75.3 92.2 7.66 Bn 90 FPS Darknet19 74.0 91.8 5.58 Bn 200 FPS
26. Why is it fast? Simple & efficient architecture C implementation 26
27. Stronger - Detecting more classes 27
28. - 14 million images - 22k classes - Classification labels - 100k images - 80 classes - Detection labels Golden eagle
29. Typically use softmax over all classes
30. Can’t just mash classes together...
31. Can’t just mash classes together...
32. WordNet has structure but it’s messy
35. ... Each node is a conditional probability
36. ... Each node is a conditional probability P(Bedlington terrier) = P(object) * P (living thing | object) * ….. P(canine | mammal) * P(dog | canine) * P(terrier | dog) * P(Bedlington terrier | terrier)
47. Conclusion • YOLOv2 and YOLO9000 real-time detection systems • YOLOv2 state of the art and faster than other systems • 9K object category detection by YOLO9000 47
48. 1. CVPR paper - https://pjreddie.com/media/files/papers/YOLO9000.pdf 2. Article - https://medium.com/@jonathan_hui/real-time-object-detection-with-yolo-yolov2-28b1b93e2088 3. Author’s Presentation - https://docs.google.com/presentation/d/14qBAiyhMOFl_wZW4dA1CkixgXwf0zKGbpw_0oHK8yEM/edit#slide=id.g1f9fb98e4b_0 _132 References 48