Analyzing the Performance of Multilayer Neural Networks for Object Recognition
arXiv:1407.1610
Abstract
In the last two years, convolutional neural networks (CNNs) have achieved an impressive suite of results on standard recognition datasets and tasks. CNN-based features seem poised to quickly replace engineered representations, such as SIFT and HOG. However, compared to SIFT and HOG, we understand much less about the nature of the features learned by large CNNs. In this paper, we experimentally probe several aspects of CNN feature learning in an attempt to help practitioners gain useful, evidence-backed intuitions about how to apply CNNs to computer vision problems.
arXiv commentPublished in European Conference on Computer Vision 2014 (ECCV-2014)
References in corpus (2)
Cited by in corpus (44)
- Particular object retrieval with integral max-pooling of CNN activations
- What makes ImageNet good for transfer learning?
- Revisiting Unreasonable Effectiveness of Data in Deep Learning Era
- Unsupervised Visual Representation Learning by Context Prediction
- Deep Learning for Generic Object Detection: A Survey
- On the importance of single directions for generalization
- BoxSup: Exploiting Bounding Boxes to Supervise Convolutional Networks for Semantic Segmentation
- Learning to See by Moving
- The Devil is in the Tails: Fine-grained Classification in the Wild
- DeepSkeleton: Learning Multi-task Scale-associated Deep Side Outputs for Object Skeleton Extraction in Natural Images
- What Do We Understand About Convolutional Networks?
- Rethinking ImageNet Pre-training
- PixelNet: Representation of the pixels, by the pixels, and for the pixels
- Towards Interpretable Deep Neural Networks by Leveraging Adversarial Examples
- Domain Adaptive Transfer Learning with Specialist Models
- How Far are We from Solving Pedestrian Detection?
- Webly Supervised Learning of Convolutional Networks
- Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks
- InLoc: Indoor Visual Localization with Dense Matching and View Synthesis
- Semantic Hierarchy Emerges in Deep Generative Representations for Scene Synthesis
- Active and Continuous Exploration with Deep Neural Networks and Expected Model Output Changes
- Unsupervised learning of object semantic parts from internal states of CNNs by population encoding
- Learning Analysis-by-Synthesis for 6D Pose Estimation in RGB-D Images
- Factors of Transferability for a Generic ConvNet Representation
- Object Skeleton Extraction in Natural Images by Fusing Scale-associated Deep Side Outputs
- Deep Learning for Single-View Instance Recognition
- Hebbian Semi-Supervised Learning in a Sample Efficiency Setting
- Object Detection Free Instance Segmentation With Labeling Transformations
- Visual Discovery at Pinterest
- AttentionNet: Aggregating Weak Directions for Accurate Object Detection
- Efficient end-to-end learning for quantizable representations
- Learning a Discriminative Filter Bank within a CNN for Fine-grained Recognition
- Deep Function Machines: Generalized Neural Networks for Topological Layer Expression
- All-Transfer Learning for Deep Neural Networks and its Application to Sepsis Classification
- The Treasure beneath Convolutional Layers: Cross-convolutional-layer Pooling for Image Classification
- Learning Sparse, Distributed Representations using the Hebbian Principle
- Brand > Logo: Visual Analysis of Fashion Brands
- Learning Models for Actions and Person-Object Interactions with Transfer to Question Answering
- Kill Two Birds With One Stone: Boosting Both Object Detection Accuracy and Speed With adaptive Patch-of-Interest Composition
- Class Subset Selection for Transfer Learning using Submodularity
- Learn Faster and Forget Slower via Fast and Stable Task Adaptation
- Manifestation of Image Contrast in Deep Networks
- Model-based active learning to detect isometric deformable objects in the wild with deep architectures
- On the Exploration of Convolutional Fusion Networks for Visual Recognition