Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Enhancing Infrared Vision: Progressive Prompt Fusion Network and Benchmark
Authors: Jinyuan Liu, Zihang Chen, Zhu Liu, Zhiying Jiang, Long Ma, Xin Fan, Risheng Liu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments substantiate that our approach not only delivers promising visual results under specific degradation but also significantly improves performance on complex degradation scenes, achieving a notable 8.76% improvement. ... 4 Experimental Results ... Training and testing data. In our experiments, we trained the TIR enhancement model on our HM-TIR dataset ... Evaluation metrics. In this work, the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) [56] are employed to assess the quality of the enhancement results... |
| Researcher Affiliation | Academia | School of Software Engineering, Dalian University of Technology Information Science and Technology College, Dalian Martime University EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1 Selective Progressive Training. Require: Clean infrared images with {Ic}, a restoration Network NĪø and other necessary hyperparameters. |
| Open Source Code | Yes | Code is available at https://github.com/Zihang-Chen/HM-TIR. |
| Open Datasets | Yes | Furthermore, we introduce the most high-quality, multi-scenarios infrared benchmark covering a wide range of scenarios. ... We establish a high-quality TIR benchmark covering multiple scenarios, named HM-TIR. ... We further evaluate our method alongside competitive approaches on the real-world Iray dataset [34] and adding an additional TIR enhancement method IE-CGAN [23]. ... Iray [34] 2021 Image 2000 256 192 horizontal ā ā§ III ... URL http://openai.raytrontek.com/apply/E_Image_noise_reduction.html/. |
| Dataset Splits | Yes | In our experiments, we trained the TIR enhancement model on our HM-TIR dataset, which contains 1,503 TIR images encompassing diverse object types across various scenarios. We divided the dataset into 80% for training and 20% for validation, ensuring a balanced evaluation of our model s performance. |
| Hardware Specification | Yes | All models are implemented in Py Torch on four 4090D GPUs with default settings." and "For fair evaluation, All the the models are equipped in hardware environment with a NVIDIA RTX 4090 D GPU with 24GB memory and the input TIR image resolution is 640 512. |
| Software Dependencies | No | All models are implemented in Py Torch on four 4090D GPUs with default settings. |
| Experiment Setup | Yes | During training, we adopt the L1 loss [56] and employ the Adam optimizer with parameters Ī²1 = 0.9 and Ī²2 = 0.999. Each model is trained with a batch size of 4, using random cropping and flipping with a patch size of 256 256. The initial learning rate is set to 8 10 5 and decays to 10 6 following a cosine annealing schedule. Each model is trained for a total of 300 epochs. |