DDR: Exploiting Deep Degradation Response as Flexible Image Descriptor
Authors: Juncheng Wu, Zhangkai Ni, Hanli Wang, Wenhan Yang, Yuyin Zhou, Shiqi Wang
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive evaluations demonstrate the versatility of DDR as an image descriptor, with strong correlations observed with key image attributes such as complexity, colorfulness, sharpness, and overall quality. Moreover, we demonstrate the efficacy of DDR across a spectrum of applications. |
| Researcher Affiliation | Academia | 1 School of Computer Science and Technology, Tongji University, China 2 Pengcheng Laboratory, China 3 Department of Computer Science and Engineering, University of California, Santa Cruz, USA 4 Department of Computer Science, City University of Hong Kong, Hong Kong |
| Pseudocode | No | No pseudocode or algorithm block found. |
| Open Source Code | Yes | Our code is available at: https://github.com/eezkni/DDR |
| Open Datasets | Yes | For image deblurring, we train and test the model using the Go Pro dataset [51] and Real Blur dataset [52], respectively. ... For SISR, we combine two real-world datasets together for training and testing, including the Real SR [53] and City100 [54] datasets. |
| Dataset Splits | No | For image deblurring, we train and test the model using the Go Pro dataset [51] and Real Blur dataset [52], respectively. (No explicit mention of validation splits). |
| Hardware Specification | Yes | All experiments are conducted using one NVIDIA RTX 4090. |
| Software Dependencies | No | For all experiment on image restoration, we employ Adam W [66] optimizer and set the batch size as 4. We train NAFNet [62] and Restormer [63] for 200,000 and 300,000 steps respectively, following their corresponding official settings. ... We use the CLIP [37] Vi T-B/32 model as the image feature extractor... (Mentions software but lacks specific version numbers). |
| Experiment Setup | Yes | For all experiment on image restoration, we employ Adam W [66] optimizer and set the batch size as 4. We train NAFNet [62] and Restormer [63] for 200,000 and 300,000 steps respectively, following their corresponding official settings. Specifically, the initial learning rate for NAFNet is set to 1e 3, and for Restormer, it is set to 3e 4. We also adopted a cosine annealing strategy for both models. We empirically train the model at a resolution of 128 x 128. |