High-Resolution Iterative Feedback Network for Camouflaged Object Detection
Authors: Xiaobin Hu, Shuo Wang, Xuebin Qin, Hang Dai, Wenqi Ren, Donghao Luo, Ying Tai, Ling Shao
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on four challenging datasets demonstrate that our Hit Net breaks the performance bottleneck and achieves significant improvements compared with 35 state-of-the-art methods. |
| Researcher Affiliation | Collaboration | Xiaobin Hu1, Shuo Wang2, Xuebin Qin3, Hang Dai4*, Wenqi Ren5, Donghao Luo1, Ying Tai1, Ling Shao6 1Tencent Youtu Lab 2ETH Zurich 3Mohamed bin Zayed University of Artificial Intelligence 4University of Glasgow 5Sun Yat-sen University 6Terminus Group {xiaobinhu, michaelluo, yingtai}@tencent.com, shawnwang.tech@gmail.com, xuebin@ualberta.ca, Hang.Dai@glasgow.ac.uk, renwq3@mail.sysu.edu.cn, ling.shao@ieee.org |
| Pseudocode | No | The paper does not contain any explicit pseudocode or algorithm blocks. |
| Open Source Code | No | Code will be made publicly available. This statement indicates a future release and does not provide concrete access at the time of publication. |
| Open Datasets | Yes | Our experiments are based on four widely-used COD datasets: (1) CHAMELEON (Skurowski et al. 2018) collects 76 high-resolution images from the Internet with the label of camouflaged animals. (2) CAMO (Le et al. 2019) includes 2,500 images with eight categories. (3) COD10K (Fan et al. 2020a) is the largest collection containing 10,000 images that divided into 10 super-classes and 78 sub-classes from multiple photography websites. (4) NC4K (Lv et al. 2021) consists of 4,121 images and is commonly used to evaluate the generalization ability of models. |
| Dataset Splits | Yes | Following previous studies and benchmarks (Zhai et al. 2021; Fan et al. 2020a; Yang et al. 2021; Lv et al. 2021), the training set includes 1,000 images from CAMO, and 3,040 images from COD10K. The test set consists of 76 images from CHAMELEON, 250 images from CAMO, 2,026 images from COD10K, and 4,121 images from NC4K. |
| Hardware Specification | Yes | We implement our model based on PyTorch in AMD Ryzen Threadripper 3990X 2.9GHz CPU and NVIDIA RTX A6000 GPU. |
| Software Dependencies | No | The paper mentions 'PyTorch' but does not specify a version number for it or any other software libraries or solvers. |
| Experiment Setup | Yes | We employ the Adam W (Loshchilov and Hutter 2019) optimizer with the learning rate of 1e-4, which is widely used in transformer structure, and the corresponding decay rate to 0.1 for every 30 epochs. The weight w of iterative feedback loss is 0.2, and the well-optimized iteration number (N) is 4. The total epochs of training are 100 with a batch size of 16. |