Balanced Learning for Domain Adaptive Semantic Segmentation
Authors: Wangkai Li, Rui Sun, Bohao Liao, Zhaoyang Li, Tianzhu Zhang
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on two standard UDA semantic segmentation benchmarks demonstrate that BLDA consistently improves performance, especially for under-predicted classes, when integrated into various existing methods. Code is available at https://github.com/Woof6/BLDA. (Section 4. Experiments) |
| Researcher Affiliation | Academia | 1Mo E Key Laboratory of Brain-inspired Intelligent Perception and Cognition, University of Science and Technology of China 2Deep Space Exploration Laboratory. Correspondence to: Tianzhu Zhang <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 Online Logits Adjustment for UDA (Appendix D) |
| Open Source Code | Yes | Code is available at https://github.com/Woof6/BLDA. |
| Open Datasets | Yes | Specifically, we use GTAv/SYNTHIA (Ros et al., 2016; Richter et al., 2016) as the labeled source domain and Cityscapes (Cordts et al., 2016) as the unlabeled target domain. |
| Dataset Splits | No | The paper states: "Following standard UDA protocols, we evaluate our method on two widely used benchmarks that involve transferring knowledge from a synthetic domain to a real domain in a street scene setting. Specifically, we use GTAv/SYNTHIA (Ros et al., 2016; Richter et al., 2016) as the labeled source domain and Cityscapes (Cordts et al., 2016) as the unlabeled target domain." While these datasets have standard splits, the paper does not explicitly provide specific percentages, sample counts, or direct references to the split methodology within its text. |
| Hardware Specification | Yes | All experiments are trained for 40K iterations and a batch size of 2, with one or two RTX-3090 (24 GB memory) GPUs, depending on the complexity of used UDA frameworks. |
| Software Dependencies | No | BLDA is implemented based on MMSegmentation (Contributors, 2020). The paper mentions MMSegmentation but does not specify a version number for it or any other key software components. |
| Experiment Setup | Yes | All experiments are trained for 40K iterations and a batch size of 2, with one or two RTX-3090 (24 GB memory) GPUs, depending on the complexity of used UDA frameworks. We train the network with an Adam W optimizer with learning rates of 6 10 5 for the encoder and 6 10 4 for the decoder, a weight decay of 0.01, and linear learning rate warm-up for the first 1.5K iterations. The input images are rescaled and randomly cropped to 512 512 following the same data augmentation in DAFormer (Hoyer et al., 2022a), and the EMA coefficient for updating the teacher net is set to be 0.999. We set temperature coefficient τ = 0.1 and loss weight λ = 0.2 respectively. |