ML Use Cases at Think Big Analytics
summary
Comparison of concrete pipelines: experiments with YOLO and SSD for real-time road object detection, a custom segmentation/annotation tool plus outsourced labeling, and data augmentation (brightness, flips) that suddenly enables pretrained models to detect roads. Tracking uses a heuristic 20-frame tracker (color plus Euclidean distance) that works better than complex research implementations. A logistics app fingerprints items by ResNet/Inception feature fingerprinting (freeze mid‑layers, cosine distance) for image search. Fraud work converts transaction sequences into visual feature maps for CNNs; LSTMs give no clear gain.
AI-written summary · about
chapters (13)
- 0:00Welcome and introduction
- 1:45Meet the guests
- 3:33Backgrounds and roles
- 5:43Tutorial on image analysis
- 7:08Challenges in training models
- 13:08Deploying models in production
- 17:21Traffic project use case
- 21:18Object detection methods
- 23:25Object tracking methods
- 27:21Traffic project insights
- 28:05Lost and found logistics project
- 35:27Bank fraud detection project
- 41:24Conference impressions
related talks
36mObject detection to Instance segmentation: Learn to apply several algorithms...
25mBridging Gaps: Classic Approaches Beyond Deep Learning Constraints
35mLarge-Scale Video Intelligence
33mRunning Machine Learning in Production - a Journey to Success
12mCascade models in computer vision to boost accuracy and performance
32mApplying Deep Learning for Object Detection from a Telecom Perspective
32mEverything you wanted to know about VLMs but were afraid to ask
41mKEYNOTE: Image processing, AI, and autonomous systems