DeepID-Net: multi-stage and deformable deep convolutional neural networks for object detection

Abstract

In this paper, we propose multi-stage and deformable deep convolutional neural networks for object detection. This new deep learning object detection diagram has innovations in multiple aspects. In the proposed new deep architecture, a new deformation constrained pooling (def-pooling) layer models the deformation of object parts with geometric constraint and penalty. With the proposed multi-stage training strategy, multiple classifiers are jointly optimized to process samples at different difficulty levels. A new pre-training strategy is proposed to learn feature representations more suitable for the object detection task and with good generalization capability. By changing the net structures, training strategies, adding and removing some key components in the detection pipeline, a set of models with large diversity are obtained, which significantly improves the effectiveness of modeling averaging. The proposed approach ranked \#2 in ILSVRC 2014. It improves the mean averaged precision obtained by RCNN, which is the state-of-the-art of object detection, from $31\%$ to $45\%$. Detailed component-wise analysis is also provided through extensive experimental evaluation.

References in corpus (6)

• Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
• OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks
• Visualizing and Understanding Convolutional Networks
• Sum-Product Networks: A New Deep Architecture
• Generic Object Detection With Dense Neural Patterns and Regionlets
• Multi-scale Orderless Pooling of Deep Convolutional Activation Features

Cited by in corpus (20)

• Training Deep Neural Networks on Noisy Labels with Bootstrapping
• DenseBox: Unifying Landmark Localization with End to End Object Detection
• Scalable, High-Quality Object Detection
• Recent Advances in Convolutional Neural Networks
• Weakly Supervised Fine-Grained Image Categorization
• Hybrid CNN and Dictionary-Based Models for Scene Recognition and Domain Adaptation
• Image Super-Resolution Using Deep Convolutional Networks
• CT-image Super Resolution Using 3D Convolutional Neural Network
• Learning Contextual Dependencies with Convolutional Hierarchical Recurrent Neural Networks
• A Discriminative Representation of Convolutional Features for Indoor Scene Recognition
• Learning like a Child: Fast Novel Visual Concept Learning from Sentence Descriptions of Images
• Multimodal Convolutional Neural Networks for Matching Image and Sentence
• Untangling Local and Global Deformations in Deep Convolutional Networks for Image Classification and Sliding Window Detection
• Boosting Optical Character Recognition: A Super-Resolution Approach
• Relation Distillation Networks for Video Object Detection
• AttentionNet: Aggregating Weak Directions for Accurate Object Detection
• Window-Object Relationship Guided Representation Learning for Generic Object Detections
• Action-Driven Object Detection with Top-Down Visual Attentions
• On-Device Text Image Super Resolution
• Temporally Identity-Aware SSD with Attentional LSTM