# Multi-scale recognition with DAG-CNNs

arXiv: https://arxiv.org/abs/1505.05232 · HTML: https://mlsanity.com/paper/1505.05232/

## Key facts

- Cited by 23 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 20 May 2015.

## Built on by (most cited first, 23 of 23)

- [CrossViT: Cross-Attention Multi-Scale Vision Transformer for Image Classification](https://mlsanity.com/paper/2103.14899/) (2021)
- [OpenGAN: Open-Set Recognition via Open Data Generation](https://mlsanity.com/paper/2104.02939/) (2021)
- [HiFormer: Hierarchical Multi-scale Representations Using Transformers for Medical Image Segmentation](https://mlsanity.com/paper/2207.08518/) (2022)
- [PCRLv2: A Unified Visual Information Preservation Framework for Self-supervised Pre-training in Medical Image Analysis](https://mlsanity.com/paper/2301.00772/) (2023)
- [Visual Attention Methods in Deep Learning: An In-Depth Survey](https://mlsanity.com/paper/2204.07756/) (2022)
- [Joint Visual Semantic Reasoning: Multi-Stage Decoder for Text Recognition](https://mlsanity.com/paper/2107.12090/) (2021)
- [Deep Learning for Scene Classification: A Survey](https://mlsanity.com/paper/2101.10531/) (2021)
- [Probabilistic Graph Attention Network with Conditional Kernels for Pixel-Wise Prediction](https://mlsanity.com/paper/2101.02843/) (2021)
- [Multi-layered Semantic Representation Network for Multi-label Image Classification](https://mlsanity.com/paper/2106.11596/) (2021)
- [MRC-LSTM: A Hybrid Approach of Multi-scale Residual CNN and LSTM to Predict Bitcoin Price](https://mlsanity.com/paper/2105.00707/) (2021)
- [SRRM: Semantic Region Relation Model for Indoor Scene Recognition](https://mlsanity.com/paper/2305.08540/) (2023)
- [ThreshNet: An Efficient DenseNet Using Threshold Mechanism to Reduce Connections](https://mlsanity.com/paper/2201.03013/) (2022)
- [Leveraging Self-Supervised Learning for Scene Classification in Child Sexual Abuse Imagery](https://mlsanity.com/paper/2403.01183/) (2024)
- [Target Detection and Segmentation in Circular-Scan Synthetic-Aperture-Sonar Images using Semi-Supervised Convolutional Encoder-Decoders](https://mlsanity.com/paper/2101.03603/) (2021)
- [New Pruning Method Based on DenseNet Network for Image Classification](https://mlsanity.com/paper/2108.12604/) (2021)
- [Recent Advances in Scene Image Representation and Classification](https://mlsanity.com/paper/2206.07326/) (2022)
- [EnTri: Ensemble Learning with Tri-level Representations for Explainable Scene Recognition](https://mlsanity.com/paper/2307.12442/) (2023)
- [Multi-task Prompt Words Learning for Social Media Content Generation](https://mlsanity.com/paper/2407.07771/) (2024)
- [Scene Retrieval for Contextual Visual Mapping](https://mlsanity.com/paper/2102.12728/) (2021)
- [Scene Recognition with Objectness, Attribute and Category Learning](https://mlsanity.com/paper/2207.10174/) (2022)
- [Look Closer to Your Enemy: Learning to Attack via Teacher-Student Mimicking](https://mlsanity.com/paper/2207.13381/) (2022)
- [Predicting Topological Maps for Visual Navigation in Unexplored Environments](https://mlsanity.com/paper/2211.12649/) (2022)
- [Intuitionistic Fuzzy Cognitive Maps for Interpretable Image Classification](https://mlsanity.com/paper/2408.03745/) (2024)

## Abstract (arXiv)

We explore multi-scale convolutional neural nets (CNNs) for image classification. Contemporary approaches extract features from a single output layer. By extracting features from multiple layers, one can simultaneously reason about high, mid, and low-level features during classification. The resulting multi-scale architecture can itself be seen as a feed-forward model that is structured as a directed acyclic graph (DAG-CNNs). We use DAG-CNNs to learn a set of multiscale features that can be effectively shared between coarse and fine-grained classification tasks. While fine-tuning such models helps performance, we show that even "off-the-self" multiscale features perform quite well. We present extensive analysis and demonstrate state-of-the-art classification performance on three standard scene benchmarks (SUN397, MIT67, and Scene15). In terms of the heavily benchmarked MIT67 and Scene15 datasets, our results reduce the lowest previously-reported error by 23.9% and 9.5%, respectively.