Relaxed Class-consensus Consistency for Semi-supervised Semantic Segmentation
Authors: Huayu Mai, Rui Sun, Feng Wu
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results on multiple benchmarks demonstrate that RCC performs favorably against state-of-the-art methods. Particularly in the low-data regimes, RCC achieves significant improvements. (...) Extensive experiments on three challenging benchmarks demonstrate that our RCC outperforms state-of-the-art semi-supervised semantic segmentation methods. |
| Researcher Affiliation | Academia | Huayu Mai , Rui Sun*, Feng Wu Deep Space Exploration Laboratory/School of Information Science and Technology, University of Science and Technology of China EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1: Pseudo algorithms of RCC. |
| Open Source Code | No | The paper does not contain an explicit statement about releasing code or a link to a code repository for the methodology described. |
| Open Datasets | Yes | Datasets. (1) PASCAL VOC 2012 (Everingham et al. 2010) is an object-centric semantic segmentation dataset... (2) Cityscapes (Cordts et al. 2016) is an urban scene understanding dataset... |
| Dataset Splits | Yes | Table 1: Quantitative results of different SSL methods on Pascal classic set. We report m Io U (%) under various partition protocols... 1/16(92) 1/8(183) 1/4(366) 1/2(732) Full(1464) (...) Table 2: ...1/16(662) 1/8(1323) 1/4(2646) (...) Table 3: ...1/16(186) 1/8(372) 1/4(744) 1/2(1488) |
| Hardware Specification | Yes | The model is trained for 80 epochs on PASCAL and 240 epochs on Cityscapes with a batch size of 8, using 8 RTX 3090 GPUs. |
| Software Dependencies | No | The paper mentions software components and techniques such as 'Res Net-50/101', 'Deep Labv3+', 'stochastic gradient descent (SGD) optimizer', and 'Polynomial Decay learning rate policy', but does not specify version numbers for any programming languages, libraries, or other software dependencies. |
| Experiment Setup | Yes | The crop size is set as 513 x 513 for PASCAL and 801 x 801 for Cityscapes, respectively. We adopt stochastic gradient descent (SGD) optimizer with an initial learning rate of 0.001 for PASCAL and 0.005 for Cityscapes. Polynomial Decay learning rate policy is applied throughout the whole training. The strong augmentation Aug( ) contains random color jitter, grayscale and Gaussian blur. The weak augmentation aug( ) consists of random crop, resize and horizontal flip. ... We set the trade-off weight λ = 0.2 for all experiments. The model is trained for 80 epochs on PASCAL and 240 epochs on Cityscapes with a batch size of 8... We take ablation experiments on N (Table 7), showing the best trade-off between performance and cost is achieved with N = 4. |