Learning Real-time Object Detection In The Absence of Large-scale Datasets
summary
Explains single-shot multibox detection (SSD) for real-time object detection, points out its limitations when labeled data is scarce, and describes extensions that combine unsupervised learning and feature-map confluence to address those shortcomings. Compares performance of the original and extended methods across multiple datasets, and presents a detailed experimental case study and practical guidance for applying object detection beyond common open-source datasets.
AI-written summary · about
chapters (13)
- 0:00Start
- 3:56Applications Of Object Detection
- 4:37Open Source Detection Datasets
- 6:10Open Source Data Is Very Noisy
- 7:21Beyond MSCOCO, Pascal, ImageNet
- 10:44Representations Via Filters
- 11:07Layers Representing Prev. Layers
- 11:40Classification Of Input Image
- 13:50Two Shot Object Detectors
- 14:57Step 1: Where To Look?
- 17:18Step 2: What Is It?
- 30:22History of Single-Shot Object Detection
- 32:21Statistical Learning? Or Brain-inspired Deep Learning? Or Reasoning? Or Artificial Intelligence?
related talks
01h 15mTrain Object Detection with Small Datasets
46mML Use Cases at Think Big Analytics
36mObject detection to Instance segmentation: Learn to apply several algorithms...
23mBuilding high-quality computer vision models using only a few examples
01h 11mObject detection using OpenCV
32mReal-Time Object Detection with CoreML
01h 09mFrom Walruses to Code: A Guide to Tackling Unique ML Problems
38mPractical Few-Shot Learning