Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper tackles synthetic-to-real transfer for tree instance segmentation, addressing the common forestry perception problem where synthetic data has fine-grained trunk/crown labels while real data only has coarse tree labels. They introduce a mixed-granularity dataset (MGTD) and a four-stage evaluation protocol to isolate the effects of domain shift and label granularity mismatch. The core contribution is a granularity-aware distillation method that transfers structural priors from synthetic teachers to real-data students using logit-space merging and mask unification, achieving improved mask AP, particularly for small or distant trees.
Bridging the synthetic-to-real gap in forestry perception doesn't require perfect labels: distilling knowledge from fine-grained synthetic data to coarse real-world labels significantly boosts tree instance segmentation, especially for hard-to-detect trees.
We address the challenge of synthetic-to-real transfer in forestry perception where real data have only coarse Tree labels while synthetic data provide fine-grained trunk/crown annotations. We introduce MGTD, a mixed-granularity dataset with 53k synthetic and 3.6k real images, and a four-stage protocol isolating domain shift and granularity mismatch. Our core contribution is granularity-aware distillation, which transfers structural priors from fine-grained synthetic teachers to a coarse-label student via logit-space merging and mask unification. Experiments show consistent mask AP gains, especially for small/distant trees, establishing a testbed for Sim-Real transfer under label granularity constraints.