# Track and Transfer: Watching Videos to Simulate Strong Human Supervision for Weakly-Supervised Object Detection

arXiv: https://arxiv.org/abs/1604.05766 · HTML: https://mlsanity.com/paper/1604.05766/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 19 Apr 2016.

## Built on by (most cited first, 9 of 9)

- [Weakly Supervised Object Localization and Detection: A Survey](https://mlsanity.com/paper/2104.07918/) (2021)
- [3D Spatial Recognition without Spatially Labeled 3D](https://mlsanity.com/paper/2105.06461/) (2021)
- [LEOD: Label-Efficient Object Detection for Event Cameras](https://mlsanity.com/paper/2311.17286/) (2023)
- [Exploiting Playbacks in Unsupervised Domain Adaptation for 3D Object Detection](https://mlsanity.com/paper/2103.14198/) (2021)
- [An Ensemble Approach for Multiple Emotion Descriptors Estimation Using Multi-task Learning](https://mlsanity.com/paper/2207.10878/) (2022)
- [Weakly Supervised Faster-RCNN+FPN to classify animals in camera trap images](https://mlsanity.com/paper/2208.14060/) (2022)
- [Discovery-and-Selection: Towards Optimal Multiple Instance Learning for Weakly Supervised Object Detection](https://mlsanity.com/paper/2110.09060/) (2021)
- [Weakly-Supervised Action Detection Guided by Audio Narration](https://mlsanity.com/paper/2205.05895/) (2022)
- [Self-Supervised 3D Monocular Object Detection by Recycling Bounding Boxes](https://mlsanity.com/paper/2206.12738/) (2022)

## Abstract (arXiv)

The status quo approach to training object detectors requires expensive bounding box annotations. Our framework takes a markedly different direction: we transfer tracked object boxes from weakly-labeled videos to weakly-labeled images to automatically generate pseudo ground-truth boxes, which replace manually annotated bounding boxes. We first mine discriminative regions in the weakly-labeled image collection that frequently/rarely appear in the positive/negative images. We then match those regions to videos and retrieve the corresponding tracked object boxes. Finally, we design a hough transform algorithm to vote for the best box to serve as the pseudo GT for each image, and use them to train an object detector. Together, these lead to state-of-the-art weakly-supervised detection results on the PASCAL 2007 and 2010 datasets.