# On the Limitation of Convolutional Neural Networks in Recognizing Negative Images

arXiv: https://arxiv.org/abs/1703.06857 · HTML: https://mlsanity.com/paper/1703.06857/

## Key facts

- Cited by 17 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 20 Mar 2017.

## Built on by (most cited first, 17 of 17)

- [Human alignment of neural network representations](https://mlsanity.com/paper/2211.01201/) (2022)
- [Robust Implicit Networks via Non-Euclidean Contractions](https://mlsanity.com/paper/2106.03194/) (2021)
- [LLM Whisperer: An Inconspicuous Attack to Bias LLM Responses](https://mlsanity.com/paper/2406.04755/) (2024)
- [Gradient-Based Adversarial and Out-of-Distribution Detection](https://mlsanity.com/paper/2206.08255/) (2022)
- [Probing the Purview of Neural Networks via Gradient Analysis](https://mlsanity.com/paper/2304.02834/) (2023)
- [Adversarial Framework with Certified Robustness for Time-Series Domain via Statistical Features](https://mlsanity.com/paper/2207.04307/) (2022)
- [Dynamic Time Warping based Adversarial Framework for Time-Series Domain](https://mlsanity.com/paper/2207.04308/) (2022)
- [XOXO: Stealthy Cross-Origin Context Poisoning Attacks against AI Coding Assistants](https://mlsanity.com/paper/2503.14281/) (2025)
- [Modular Multimodal Machine Learning for Extraction of Theorems and Proofs in Long Scientific Documents (Extended Version)](https://mlsanity.com/paper/2307.09047/) (2023)
- [Pulmonary Disease Classification Using Globally Correlated Maximum Likelihood: an Auxiliary Attention mechanism for Convolutional Neural Networks](https://mlsanity.com/paper/2109.00573/) (2021)
- [Denoised Internal Models: a Brain-Inspired Autoencoder against Adversarial Attacks](https://mlsanity.com/paper/2111.10844/) (2021)
- [Training Robust Deep Models for Time-Series Domain: Novel Algorithms and Theoretical Analysis](https://mlsanity.com/paper/2207.04305/) (2022)
- [Group-based Robustness: A General Framework for Customized Robustness in the Real World](https://mlsanity.com/paper/2306.16614/) (2023)
- [Analyzing Adversarial Robustness of Deep Neural Networks in Pixel Space: a Semantic Perspective](https://mlsanity.com/paper/2106.09872/) (2021)
- [Attack-Centric Approach for Evaluating Transferability of Adversarial Samples in Machine Learning Models](https://mlsanity.com/paper/2112.01777/) (2021)
- [On the ability of CNNs to extract color invariant intensity based features for image classification](https://mlsanity.com/paper/2307.06500/) (2023)
- [Towards In-Vehicle Multi-Task Facial Attribute Recognition: Investigating Synthetic Data and Vision Foundation Models](https://mlsanity.com/paper/2403.06088/) (2024)

## Abstract (arXiv)

Convolutional Neural Networks (CNNs) have achieved state-of-the-art performance on a variety of computer vision tasks, particularly visual classification problems, where new algorithms reported to achieve or even surpass the human performance. In this paper, we examine whether CNNs are capable of learning the semantics of training data. To this end, we evaluate CNNs on negative images, since they share the same structure and semantics as regular images and humans can classify them correctly. Our experimental results indicate that when training on regular images and testing on negative images, the model accuracy is significantly lower than when it is tested on regular images. This leads us to the conjecture that current training methods do not effectively train models to generalize the concepts. We then introduce the notion of semantic adversarial examples - transformed inputs that semantically represent the same objects, but the model does not classify them correctly - and present negative images as one class of such inputs.