Object detection to Instance segmentation: Learn to apply several algorithms...
summary
Compares Faster R‑CNN, SSD, YOLO and an open‑source Mask R-CNN pipeline that uses ResNet‑101/feature pyramid backbones, an RPN, and a mask branch trained with a combined classification/regression/mask loss. Notes ROIAlign reduces mask and box error, COCO has 80 classes so many projects convert labels to COCO-style. Describes building a prototype on the balloon dataset with labelme/VoTT and COCO tools that works for medium objects but misses heavy occlusion and very small items; YOLOv3 trades accuracy for speed unless a larger backbone is used.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:31Computer vision fundamentals
- 5:01Object detection overview
- 10:21YOLO evolution
- 14:56Detection model comparison
- 15:16Instance segmentation intro
- 17:01Mask R-CNN architecture
- 21:23Industry applications
- 23:30Data prep for Mask R-CNN
- 26:01Real-world challenges
- 27:11Latest research limits
- 28:20Demo walkthrough
- 30:37Q&A
related talks
01h 15mTrain Object Detection with Small Datasets
01h 29mHuman-like Visual Search Application with Small Data
46mML Use Cases at Think Big Analytics
01h 07mComputer Vision with PyTorch
12mCascade models in computer vision to boost accuracy and performance
01h 11mObject detection using OpenCV
22mObject Detection with KerasCV
01h 14mFoundations of Object Detection from an Expert's Overview