How fishandwasabi/yolo-ms works
This repository implements YOLO-MS, a real-time object-detection research codebase, with an MMYOLO-based model plus Segment Anything utilities for promptable and automatic mask generation. The principal workflow is to configure and train a detector, whose multi-scale backbone and feature pyramid use MSBlocks; detection heads and dataset-specific configurations are also present. Separate branches cover dataset preparation and deployment configurations. The graph reflects only relationships visible in the supplied excerpts, SOURCE INDEX, or README; unsampled model internals are not linked speculatively.
Main entry points: predictor.py Prompt predictor · msblock.py Multi-scale block · automatic_mask_generator.py Automatic masks · train.py Training entry
Components 19
Detection model
- YOLO-MS backbone
mmyolo/yoloms/models/backbones/yoloms.py - Multi-scale block
mmyolo/yoloms/models/layers/msblock.py - Block layers
mmyolo/yoloms/models/layers/msblock_layers.py - Feature pyramid
mmyolo/yoloms/models/necks/yoloms_pafpn.py - Detection head
mmyolo/yoloms/models/dense_head/yoloms_head.py - Distillation detector
mmyolo/yoloms/models/detectors/yolo_distill.py
Mask generation
- Prompt predictor
mmyolo/yoloms/models/segment_anything/predictor.py - Image encoder
mmyolo/yoloms/models/segment_anything/modeling/image_encoder.py - Encoder components
mmyolo/yoloms/models/segment_anything/modeling/common.py - SAM model components
mmyolo/yoloms/models/segment_anything/modeling - Automatic masks
mmyolo/yoloms/models/segment_anything/automatic_mask_generator.py - Mask utilities
mmyolo/yoloms/models/segment_anything/utils/amg.py - Image transforms
mmyolo/yoloms/models/segment_anything/utils/transforms.py
Training and data
- Training entry
mmyolo/tools/train.py - MMEngine runner
- Model configurations
mmyolo/configs/yoloms - Dataset tools
mmyolo/tools/dataset_converters
Deployment
- Deployment configs
mmyolo/configs/deployment
Other components
- Researcher
Connections 10
- Researcher → Training entry: starts training
mmyolo/tools/train.py - Training entry → MMEngine runner: builds and runs
mmyolo/tools/train.py - YOLO-MS backbone → Multi-scale block: uses blocks
mmyolo/yoloms/models/backbones/yoloms.py - Multi-scale block → Block layers: builds layers
mmyolo/yoloms/models/layers/msblock.py - Feature pyramid → Multi-scale block: uses blocks
mmyolo/yoloms/models/necks/yoloms_pafpn.py - Prompt predictor → SAM model components: imports model
mmyolo/yoloms/models/segment_anything/predictor.py - Prompt predictor → Image transforms: uses transforms
mmyolo/yoloms/models/segment_anything/predictor.py - Image encoder → Encoder components: uses components
mmyolo/yoloms/models/segment_anything/modeling/image_encoder.py - Automatic masks → Prompt predictor: creates predictor
mmyolo/yoloms/models/segment_anything/automatic_mask_generator.py - Automatic masks → Mask utilities: uses utilities
mmyolo/yoloms/models/segment_anything/automatic_mask_generator.py