TSHRSeg-Cane review weights

This public model repository provides checkpoints for the comparison experiments and the TSHRSeg-Cane-IDR two-stage review workflow. The corresponding anonymous source package contains three representative 4K images, ground-truth annotations, a locked environment, automatic download/verification code, and executable inference, evaluation, and benchmark commands.

Released checkpoints

Method Path(s) Inference setting
YOLOv8n-Seg (1024) Model/v8n_1024_best.pt 1024-pixel input
YOLOv8n-P6-Seg (1024) Model/v8n-p6_1024_best.pt 1024-pixel input
YOLOv8n-Seg (3840) Model/v8n_3840_best.pt 3840-pixel input
YOLOv8n-P6-Seg (3840) Model/v8n-p6_3840_best.pt 3840-pixel input
ISDNet three checkpoint paths under Model/ISDNet/checkpoints/ Three class-specific 3840 × 2160 models
CascadePSP Model/CascadePSP/checkpoints/model_45000.pth plus the 1024-pixel YOLO checkpoint Class-wise 4K refinement
TSHRSeg-Cane-IDR Model/TSHRSeg-Cane-IDR/OBB_best.pt; Model/TSHRSeg-Cane-IDR/CARR_without_correction_best.pt 640-pixel OBB stage and 1024 × 1024 reorganized-ROI stage without angle correction

The class order is bud_wing, bud, and internode. Predictions are mapped to the coordinate system of the original image before visualization or metric calculation.

Intentionally withheld during peer review

The complete TSHRSeg-Cane checkpoint and TSHRSeg-Cane-IDS checkpoint will be added after peer review. The released TSHRSeg-Cane-IDR weights are supplied specifically for framework testing; they are not the final TSHRSeg-Cane weights. They allow reviewers to execute and inspect a complete two-stage localization, ROI-reorganization, segmentation, and inverse-mapping workflow. Under the unified evaluation reported in the manuscript, this IDR configuration already exceeds the corresponding single-stage baseline, while the final angle-corrected configuration remains withheld until the post-review release.

Download

From the source-repository root:

huggingface-cli download Arvin8888888/TSHRSeg-Cane-baseline-weights --local-dir . --local-dir-use-symlinks False

The source launchers download missing released files automatically, so this manual step is a fallback. Keep the Model/ hierarchy unchanged. Model/weights_manifest.json and Model/SHA256SUMS.txt provide file-integrity metadata.

Intended use and limitations

These checkpoints are intended for research reproduction and comparative evaluation of fine-grained segmentation in high-resolution sugarcane stem images. Performance outside the acquisition conditions and class definitions described in the associated manuscript has not been established. They should not be used as the sole basis for breeding, agronomic, or commercial decisions.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support