Synonymous Variational Inference for Perceptual Image Compression

Authors: Zijian Liang, Kai Niu, Changshuo Wang, Jin Xu, Ping Zhang

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate comparable rate-distortion-perception performance using a single progressive SIC codec, thus verifying the effectiveness of our proposed analysis method.
Researcher Affiliation Academia 1Key Laboratory of Universal Wireless Communications, Ministry of Education, Beijing University of Posts and Telecommunications, Beijing, China 2Peng Cheng Laboratory, Shenzhen, China 3State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China. Correspondence to: Kai Niu <EMAIL>.
Pseudocode No The paper describes methods and processes through text and diagrams (e.g., Figure 1, Figure 2, Figure 3), but it does not include a dedicated section or figure explicitly labeled as 'Pseudocode' or 'Algorithm' with structured, code-like steps.
Open Source Code Yes We will upload code for reproducing our results to the repository at https://github.com/ZJLiang6412/ Synonymous Image Compression.
Open Datasets Yes We randomly select 100,000 images from the Open Images V6 dataset (Kuznetsova et al., 2020) as the training data... We evaluate our models with the test set of CLIC2020 (Toderici et al., 2020), the validation set of DIV2K (Agustsson & Timofte, 2017), and the Kodak dataset 4. 4Kodak Photo CD dataset, URL http://r0k.us/graphics/kodak/.
Dataset Splits Yes We randomly select 100,000 images from the Open Images V6 dataset (Kuznetsova et al., 2020) as the training data... We evaluate our models with the test set of CLIC2020 (Toderici et al., 2020), the validation set of DIV2K (Agustsson & Timofte, 2017), and the Kodak dataset.
Hardware Specification No The paper mentions training models and performing experiments but does not specify any particular hardware components such as GPU models, CPU types, or memory sizes used for these computations.
Software Dependencies No The paper mentions implementing components using the Swin Transformer and utilizing the AdamW optimizer but does not specify version numbers for any software libraries, frameworks, or languages (e.g., Python, PyTorch, TensorFlow).
Experiment Setup Yes We train our model for 1 x 10^6 iterations with a batch size of 16, a learning rate of 1 x 10^-4, and the AdamW optimizer with a weight decay of 5 x 10^-5. Table 1 presents these hyperparameters of our progressive SIC model, which are configured empirically. We empirically set a = 4, b = 64 in this constraint.