Model Zoo

visdet’s “supported models” are the YAML model presets under configs/models/.

At runtime these are discovered automatically (via MODEL_PRESETS), so adding a new YAML file under configs/models/ makes it available everywhere that accepts a model preset name (e.g. SimpleRunner(model=...), DetInferencer(model=...)).

Using A Model Preset

Inference

from visdet.apis import DetInferencer

# Either use the preset name…
infer = DetInferencer(model="rtmdet_s")

# …or (when present) an alias from `preset_meta.aliases`
infer = DetInferencer(model="rtmdet-s")

result = infer("image.jpg")

Training

from visdet import SimpleRunner

runner = SimpleRunner(
    model="rtmdet_s",
    dataset="coco_detection",
    optimizer="adamw_default",
    scheduler="1cycle",
)
runner.train()

Supported Models (from configs/models)

Two-Stage Detectors

• faster_rcnn_r50
• mask_rcnn_r50
• mask_rcnn_swin_s

RTMDet (Object Detection)

• rtmdet_tiny (aliases: rtmdet-t, rtmdet-tiny)
• rtmdet_s (alias: rtmdet-s)
• rtmdet_m (alias: rtmdet-m)
• rtmdet_l (alias: rtmdet-l)
• rtmdet_x (alias: rtmdet-x)

RTMDet-Ins (Instance Segmentation)

• rtmdet_ins_tiny (aliases: rtmdet-ins-t, rtmdet-ins-tiny)
• rtmdet_ins_s (alias: rtmdet-ins-s)
• rtmdet_ins_m (alias: rtmdet-ins-m)
• rtmdet_ins_l (alias: rtmdet-ins-l)
• rtmdet_ins_x (alias: rtmdet-ins-x)

Discoverability

from visdet import SimpleRunner

print(SimpleRunner.list_models())

from visdet.apis import DetInferencer

print(DetInferencer.list_models())