Search papers, labs, and topics across Lattice.
This paper addresses the challenge of nighttime image dehazing under limited supervision by constructing a domain-aligned dataset using CLIP to select relevant external samples. They train a NAFNet model in two stages, first adapting to the target domain and then generalizing to broader degradation patterns. The method achieves improved dehazing performance by combining TLC, x8 self-ensemble, and weighted snapshot fusion at inference time, demonstrating a practical pipeline without complex network redesign.
Counterintuitively, leveraging CLIP to curate external data boosts nighttime image dehazing performance more effectively than complex network architectures when training data is limited.
Nighttime image dehazing faces a more complex degradation pattern than its daytime counterpart, as haze scattering couples with low illumination, non-uniform lighting, and strong light interference. Under limited supervision, this complexity aggravates domain drift and training instability, since target-domain samples are scarce while naively introducing external data may weaken adaptation due to distribution mismatch. This paper presents our solution to the NTIRE 2026 Night Time Image Dehazing Challenge, built as a unified framework that integrates domain-aligned data construction, stage-wise training, and inference-time enhancement. Specifically, a pre-trained CLIP visual encoder screens candidate external samples by similarity to construct training data closer to the target domain. NAFNet is then trained in two stages, first adapting to the target domain and then expanding to broader degradation patterns. At inference time, TLC, x8 self-ensemble, and weighted snapshot fusion are combined to improve output stability. Rather than relying on complex network redesign, the proposed framework offers a practical and effective pipeline for nighttime image dehazing.
