Technical Report for the ICRA 2026 GOOSE 2D Fine-Grained Semantic Segmentation Challenge: Pretraining-Diverse Ensemble of Foundation Vision Encoders for Robust Outdoor Scene Understanding
2026-06-22 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors describe their method for a challenge where the goal was to identify 56 detailed types of objects in outdoor scenes from multiple cameras. They combined three different vision encoders with a Mask2Former decoder and trained the system using careful techniques like longer training, image augmentation, and ensemble methods. Their approach achieved a high accuracy score, earning second place in the competition. They also found that how the encoders were pretrained mattered more for success than the number of model parameters or the decoder design.
semantic segmentationvision encoderMask2Formerpretrainingensemble learningmean IoUdata augmentationfine-grained categoriescomputer vision challengetest-time augmentation
Authors
Boyan Wang, Yongxi Huang, Wenjing Li, Tianrui Hui, Shaofei Huang, Nan Pu, Zhun Zhong
Abstract
This report presents our solution for the ICRA 2026 GOOSE 2D Fine-Grained Semantic Segmentation Challenge, which requires parsing unstructured outdoor scenes from four camera platforms into 56 fine-grained categories. Our approach pairs foundation vision encoders (including DINOv3, SigLIP2, and InternImage) with a Mask2Former decoder, and trains them with a strong recipe including long training schedules, exponential moving average, a larger crop size, and multi-scale plus flip test-time augmentation. The three encoders, chosen for their complementary pretraining objectives, are combined into a pretraining-diverse ensemble through per-class validation-IoU weighting. Evaluated on the official GOOSE test set, our submission achieves 75.40% composite mIoU and wins the second place of the challenge. Our study further shows that the encoder's pretraining recipe, rather than its parameter count or the decoder design, is the dominant factor for accuracy on this benchmark.