Multi-Modality Deep Network for JPEG Artifacts Reduction
Authors: Xuhao Jiang, Weimin Tan, Qing Lin, Chenxi Ma, Bo Yan, Liquan Shen
IJCAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments, including a user study, prove that our method can obtain better deblocking results compared to the SOTA methods. 4 Experimental Results 4.1 Datasets and Evaluation Methodology We evaluate our TGJAR on the CUB [Wah et al., 2011] and Oxford-102 [Nilsback and Zisserman, 2008] datasets, in which all images are annotated with corresponding text descriptions. 4.2 Comparison with The SOTA Methods In this part, TGJAR and the SOTA algorithms including EDSR [Lim et al., 2017], RNAN [Zhang et al., 2019], QGCN [Li et al., 2020a] and FBCNN [Jiang et al., 2021] are compared quantitatively and qualitatively. 4.3 User Study We further conduct a user study with 13 subjects. 4.4 Ablation Study We conduct the ablation study to verify the effectiveness of the proposed contrastive loss and two image-text fusion modules including GFM and LFM. |
| Researcher Affiliation | Academia | Xuhao Jiang1 , Weimin Tan1 , Qing Lin1 , Chenxi Ma1 , Bo Yan1 and Liquan Shen2 1School of Computer Science, Shanghai Key Laboratory of Intelligent Information Processing, Shanghai Collaborative Innovation Center of Intelligent Visual Computing, Fudan University, Shanghai, China 2School of Communication, Shanghai University, Shanghai, China |
| Pseudocode | No | The paper does not contain any pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statements about open-sourcing the code for their method, nor does it include a link to a code repository. |
| Open Datasets | Yes | We evaluate our TGJAR on the CUB [Wah et al., 2011] and Oxford-102 [Nilsback and Zisserman, 2008] datasets, in which all images are annotated with corresponding text descriptions. |
| Dataset Splits | No | CUB dataset contains 200 species of bird with a total of 11,788 images, of which 8,855 images are used for training and 2,933 images are used for testing. Oxford-102 Dataset consists of 102 flower categories, with a total of 8,189 images including 7,034 images for training and 1,155 images for testing. The paper provides train and test splits, but no explicit validation split. |
| Hardware Specification | Yes | All the experiments are conducted on a NVIDIA Ge Force RTX 1080 Ti. |
| Software Dependencies | No | Pytorch is used as the training toolbox, and the Adam optimization algorithm [Kingma and Ba, 2014] with a minibatch of 4 is adopted for training. The paper mentions PyTorch and Adam optimizer but does not specify their version numbers or versions for any other software libraries. |
| Experiment Setup | Yes | Pytorch is used as the training toolbox, and the Adam optimization algorithm [Kingma and Ba, 2014] with a minibatch of 4 is adopted for training. The learning rate is changed from 1 × 10−4 to 1 × 10−8 at the interval of twenty epochs. The hyper-parameter c of the LC is set as 0.1, and the hyper-parameters λ1, λ2, λ3, and λ4 of the global loss function are empirically set as 0.01, 1, 0.001 and 0.0005, respectively. |