Progressive Feature Self-Reinforcement for Weakly Supervised Semantic Segmentation
Authors: Jingxuan He, Lechao Cheng, Chaowei Fang, Zunlei Feng, Tingting Mu, Mingli Song
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments conducted on PASCAL VOC 2012 and MS COCO 2014 demonstrate that our proposed single-stage approach for WSSS not only outperforms state-of-the-art counterparts but also surpasses multi-stage methods that trade complexity for accuracy. |
| Researcher Affiliation | Academia | Jingxuan He1, Lechao Cheng1*, Chaowei Fang2, Zunlei Feng3, Tingting Mu4, Mingli Song3 1Zhejiang Lab, 2Xidian University, 3Zhejiang University, 4University of Manchester |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain an explicit statement about the release of its source code or a link to a code repository. |
| Open Datasets | Yes | We evaluate our method on two benchmarks: PASCAL VOC 2012 (Everingham et al. 2010) and MS COCO 2014 (Lin et al. 2014). |
| Dataset Splits | Yes | Following the common practice of previous works (Zhang et al. 2020; Araslanov and Roth 2020; Ru et al. 2022, 2023), it is augmented with data from the SBD dataset (Hariharan et al. 2011), resulting in 10, 582, 1, 449 and 1, 456 images for training, validation and testing, respectively. |
| Hardware Specification | No | The paper does not specify the hardware used for its experiments, such as specific GPU or CPU models. |
| Software Dependencies | No | The paper mentions several components and optimizers (e.g., ViT-B, Deep Lab-Large FOV, AdamW) and data augmentation techniques but does not specify version numbers for any software dependencies or programming languages used. |
| Experiment Setup | Yes | The base learning rate is warmed up to 6e 5 and decayed with a cosine schedule. The weighting factors (λ1, λ2, λ3, λ4, λ5) are (1.0, 0.2, 0.1, 0.1, 0.1). The momentum for EMA is 0.9995 and increases to 1.0 with a cosine schedule during training. The masking ratio r is 0.4 for adaptive uncertain feature selection. The background scores (βl, βh) introduced to determine uncertain regions are (0.2, 0.7). Training iterations are 20,000 for PASCAL VOC 2012 and 80,000 for MS COCO 2014. |