Fast and robust object detection
Fine-grained object recognition
Multi-task Deep CNN
3D human pose estimation
Person re-identifcation
Semantic attribute parsing
Real-time multiple object tracking
Detecting interactions
Activity recognition
Event detection in videos
Crowd behaviour analysis
Learning with few data
One-class learning
Self-supervised learning of image and video representations
Domain adaptation
Multimodal visual data annotation
Interactive detection and segmentation
Automatic label temporal propagation
PIXANO
Active learning
Reinforcement learning for autonomous navigation
Adversarial attacks and defense in deep metric learning
previous arrow
next arrow
Fast and robust object detection
Fast and robust object detection

The objective of an object detector is to predict a bounding box and a class for each object of interest present in the image. To be efficient, and object detector must be robust to a number of factors such as size variation and occlusion of the objects to be detected. We have developed the LapNet object detector [1], which provides a very good compromise between application performance and computation time. The learning procedure proposed in LapNet allows to efficiently manage occlusions between objects in the same image. An automatic balancing method is also proposed, allowing to balance the cost functions with respect to the representativeness of the classes and the size of the objects in the training dataset.

[1] LapNet : Automatic Balanced Loss and Optimal Assignment for Real-Time Dense Object Detection, F. Chabot, Q-C. Pham, M. Chaouch, https://arxiv.org/pdf/1911.01149.pdf, 2019

previous arrowprevious arrow
next arrownext arrow