# Human-Aligned Evaluation of a Pixel-wise DNN Color Constancy Model

arXiv: https://arxiv.org/abs/2602.13887 · HTML: https://mlsanity.com/paper/2602.13887/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 12 of its 70 references that are in the map.
- First posted on arXiv 14 Feb 2026.

## Builds on (most cited first, 12 of 12)

- [U-Net: Convolutional Networks for Biomedical Image Segmentation](https://mlsanity.com/paper/1505.04597/) (2015)
- [Cross-Camera Convolutional Color Constancy](https://mlsanity.com/paper/2011.11890/) (2020)
- [CLCC: Contrastive Learning for Color Constancy](https://mlsanity.com/paper/2106.04989/) (2021)
- [Cascading Convolutional Color Constancy](https://mlsanity.com/paper/1912.11180/) (2019)
- [Auto White-Balance Correction for Mixed-Illuminant Scenes](https://mlsanity.com/paper/2109.08750/) (2021)
- [Color Constancy Using CNNs](https://mlsanity.com/paper/1504.04548/) (2015)
- [Single and Multiple Illuminant Estimation Using Convolutional Neural Networks](https://mlsanity.com/paper/1508.00998/) (2015)
- [Attentive Illumination Decomposition Model for Multi-Illuminant White Balancing](https://mlsanity.com/paper/2402.18277/) (2024)
- [ColorSense: A Study on Color Vision in Machine Visual Recognition](https://mlsanity.com/paper/2212.08650/) (2022)
- [Divergences in Color Perception between Deep Neural Networks and Humans](https://mlsanity.com/paper/2309.05809/) (2023)
- [Pixel-Wise Color Constancy via Smoothness Techniques in Multi-Illuminant Scenes](https://mlsanity.com/paper/2402.02922/) (2024)
- [Deep chroma compression of tone-mapped images](https://mlsanity.com/paper/2409.16032/) (2024)

## Abstract (arXiv)

We previously investigated color constancy in photorealistic virtual reality (VR) and developed a Deep Neural Network (DNN) that predicts reflectance from rendered images. Here, we combine both approaches to compare and study a model and human performance with respect to established color constancy mechanisms: local surround, maximum flux and spatial mean. Rather than evaluating the model against physical ground truth, model performance was assessed using the same achromatic object selection task employed in the human experiments. The model, a ResNet based U-Net from our previous work, was pre-trained on rendered images to predict surface reflectance. We then applied transfer learning, fine-tuning only the network's decoder on images from the baseline VR condition. To parallel the human experiment, the model's output was used to perform the same achromatic object selection task across all conditions. Results show a strong correspondence between the model and human behavior. Both achieved high constancy under baseline conditions and showed similar, condition-dependent performance declines when the local surround or spatial mean color cues were removed.