Assess PO-ELIC performance beyond CLIC 2022 and at low resolutions

Ascertain the performance of the PO-ELIC image compression method on low-resolution datasets and on datasets other than CLIC 2022, given that public code or broader reconstructions are unavailable.

Background

In the comparative evaluation, the authors note that VLIC underperforms PO-ELIC on the CLIC 2022 dataset. However, PO-ELIC has not released public code and only provides reconstructions for CLIC 2022, which contains just 30 images, limiting broader assessment across resolutions and datasets.

The sentence explicitly points out that PO-ELIC’s performance at low resolutions and on other datasets remains unclear due to the lack of released code and comprehensive reconstructions, making this an unresolved evaluation question.

References

VLIC under-performs relative to PO-ELIC on CLIC 2022, but without released code, the performance of PO-ELIC on low-resolution data or other datasets is unclear, and it is important to note that CLIC 2022 contains only 30 images.

— VLIC: Vision-Language Models As Perceptual Judges for Human-Aligned Image Compression (2512.15701 - Sargent et al., 17 Dec 2025) in Section 4 (Experiments), Main results