A Well-aligned Dataset for Learning Image Signal Processing on Smartphones from a High-end Camera

SIGGRAPH 2022 Poster

Yazhou Xing1, Changlin Li1 , Xuaner Zhang2, Qifeng Chen1
1The Hong Kong University of Science and Technology, 2Adobe Inc.



Not every camera is equipped with an excellent image signal processing (ISP) pipeline that converts raw sensor data into color images. In this paper, we present a novel learning-based model that replaces built-in ISP and synthesizes images that match the image quality from high-end professional cameras. Our approach does not rely on the sub-optimal built-in ISP at all but instead utilizes a fully convolutional network with content-aware conditional convolutions to act as ISP. To train the deep learning model, we collect a large-scale dataset with raw and RGB data pairs captured by two popular smartphones and one high-end camera. Our model takes the raw sensor data from a smartphone as input and generates an RGB image that is optimized to reach the image quality coming from the high-end camera ISP. Experimental results show that our presented model produces perceptually better images than the popular smartphones do when using the same sensor data.

Our Poster


Dataset Examples

Here are two examples of our dataset. We collect our data triplet with Mi phone, iPhone 6S, and Nikon Z6.


Misalignment analysis

Misalignment analysis. In our dataset, most patches have misalignment to 0.4 ~ 0.7 pixels. The same misalignment analysis on different illuminations is consistent with overall misalignment distribution, as seen in (b).


Visual results

We compare the results of our model with smartphone build-in ISPs, and other baseline methods.

scales scales

Quantitative results

Quantitative comparison among our model and all baseline methods. Overall, all perceptual metrics show that our proposed ISP model outperforms the baselines.


Ablation study

Quantitative comparison for our controlled experiments.



  • Our dataset is available for non-commercial research purposes only.
  • You agree not to reproduce, duplicate, copy, sell, trade, resell or exploit for any commercial purposes, any portion of the videos and any portion of derived data.
  • You agree not to further copy, publish or distribute any portion of our dataset. Except, for internal use at a single site within the same organization it is allowed to make copies of the dataset.


If you find this helpful, please cite our work:

        title={A Well-aligned Dataset for Learning Image Signal Processing on Smartphones from a High-end Camera},
        author={Xing, Yazhou and Li, Changlin and Zhang, Xuaner and Chen, Qifeng},
        booktitle={ACM SIGGRAPH 2022 Posters},