CoNet++: Bilinear and Multi-Scale Feature Fusion for Real-Time Exposure Correction
DOI:
https://doi.org/10.54097/bp841717Keywords:
Exposure correction, Image enhancement;, Three-dimensional lookup table, Feature fusion, AttentionAbstract
Learnable three-dimensional lookup tables provide efficient global color transformation for exposure correction, but their pixel-value mapping is weak at spatially localized degradation. Collaborative transformation frameworks partly address this limitation by combining a lookup-table branch with a low-resolution pixel-wise branch, although simple cross-branch fusion and scale aggregation can still blur local details and misalign contextual features. We introduce CoNet++, an extension of CoTF that inserts a multi-scale feature fusion module at the encoder bottleneck and a bilinear fusion module before deep cross-branch interaction. The former aggregates block-wise local and global context with channel gating, whereas the latter jointly models spatial and channel responses for adaptive feature alignment. On the LCDP test set, CoNet++ achieves 24.1819 dB PSNR, 0.8666 SSIM, and 0.1430 LPIPS, compared with 23.9292 dB, 0.8559, and 0.1507 for the reproduced CoTF baseline. The full model contains 0.3177 million parameters and requires 0.9330 GFLOPs for the local 256-pixel profiling setting. These results show that the two modules are complementary, while the increased latency and the single-dataset evaluation define the present applicability boundary.
Downloads
References
[1] Huang, X., Zhang, Q., Hu, J.-F., & Zheng, W.-S. (2025). CLIP-RestoreX: Restore image structure and perception in exposure correction. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 39, No. 4, pp. 3760-3768). AAAI Press. https://doi.org/10.1609/aaai.v39i4.32392 DOI: https://doi.org/10.1609/aaai.v39i4.32392
[2] Huang, M., Chang, K., Qin, Q., Tang, Y., & Li, G. (2025). Conditional Laplacian pyramid networks for exposure correction. Signal Processing: Image Communication, 134, Article 117276. https://doi.org/10.1016/j.image.2025.117276 DOI: https://doi.org/10.1016/j.image.2025.117276
[3] Wang, Y., Peng, L., Li, L., Cao, Y., & Zha, Z.-J. (2023). Decoupling-and-aggregating for image exposure correction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 18115-18124). IEEE. DOI: https://doi.org/10.1109/CVPR52729.2023.01737
[4] Baek, J.-H., Kim, D., Choi, S.-M., Lee, H.-J., Kim, H., & Koh, Y. J. (2023). Luminance-aware color transform for multiple exposure correction. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) (pp. 6156-6165). IEEE. DOI: https://doi.org/10.1109/ICCV51070.2023.00566
[5] Zeng, H., Cai, J., Li, L., Cao, Z., & Zhang, L. (2022). Learning image-adaptive 3D lookup tables for high performance photo enhancement in real-time. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(4), 2058-2073. https://doi.org/10.1109/TPAMI.2020.3044140
[6] Kim, W., & Cho, N. I. (2024). Image-adaptive 3D lookup tables for real-time image enhancement with bilateral grids. In Proceedings of the European Conference on Computer Vision (ECCV). Springer. https://doi.org/10.1007/978-3-031-72967-6_6 DOI: https://doi.org/10.1007/978-3-031-72967-6_6
[7] Yang, C., Jin, M., Jia, X., Xu, Y., & Chen, Y. (2022). AdaInt: Learning adaptive intervals for 3D lookup tables on real-time image enhancement. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 17522-17531). IEEE. DOI: https://doi.org/10.1109/CVPR52688.2022.01700
[8] Li, Z., Zhang, F., Cao, M., Zhang, J., Shao, Y., Wang, Y., & Sang, N. (2024). Real-time exposure correction via collaborative transformations and adaptive sampling. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 2984-2994). IEEE. DOI: https://doi.org/10.1109/CVPR52733.2024.00288
[9] Yang, C., Jin, M., Xu, Y., Zhang, R., Chen, Y., & Liu, H. (2022). SepLUT: Separable image-adaptive lookup tables for real-time image enhancement. In Proceedings of the European Conference on Computer Vision (ECCV) (pp. 201-217). Springer. DOI: https://doi.org/10.1007/978-3-031-19797-0_12
[10] Hu, J., Shen, L., & Sun, G. (2018). Squeeze-and-excitation networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 7132-7141). IEEE. https://doi.org/10.1109/CVPR.2018.00745 DOI: https://doi.org/10.1109/CVPR.2018.00745
[11] Zamir, S. W., Arora, A., Khan, S., Hayat, M., Khan, F. S., & Yang, M.-H. (2022). Restormer: Efficient transformer for high-resolution image restoration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 5728-5739). IEEE. https://doi.org/10.1109/CVPR52688.2022.00564 DOI: https://doi.org/10.1109/CVPR52688.2022.00564
[12] Wang, Z., Bovik, A. C., Sheikh, H. R., & Simoncelli, E. P. (2004). Image quality assessment: From error visibility to structural similarity. IEEE Transactions on Image Processing, 13(4), 600-612. https://doi.org/10.1109/TIP.2003.819861 DOI: https://doi.org/10.1109/TIP.2003.819861
[13] Zhang, R., Isola, P., Efros, A. A., Shechtman, E., & Wang, O. (2018). The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 586-595). IEEE. DOI: https://doi.org/10.1109/CVPR.2018.00068
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Journal of Computing and Electronic Information Management
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
