File size: 3,938 Bytes
9356c5a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 | # Models
This directory contains pre-trained ONNX models for object detection and pose estimation.
## Directory Structure
```
models/
βββ yolo/
β βββ yolov12n.onnx # YOLO12 Nano (object detection)
β βββ yolov26n.onnx # YOLO12 Nano (alternative)
βββ rtmpose/
β βββ end2end.onnx # RTMW (pose estimation)
βββ README.md
```
## Available Models
### Object Detection
| Model | File | Size | Classes | Input | Description |
|-------|------|------|---------|-------|-------------|
| YOLO12n | `yolo/yolov12n.onnx` | ~11 MB | 80 COCO | 640Γ640 | YOLO12 Nano - fastest |
| YOLOv26n | `yolo/yolov26n.onnx` | ~11 MB | 80 COCO | 640Γ640 | YOLO12 Nano variant - balanced |
**Model Selection:**
- **YOLO12n**: Fastest inference, best for real-time applications
- **YOLv26n**: Slightly different architecture, may have better accuracy on some objects
### Pose Estimation
| Model | File | Size | Keypoints | Input | Description |
|-------|------|------|-----------|-------|-------------|
| RTMW | `rtmpose/end2end.onnx` | ~50 MB | 17 COCO | 384Γ288 | RTMW wholebody |
## Usage
### ObjectDetector
```typescript
import { ObjectDetector } from 'rtmlib-ts';
const detector = new ObjectDetector({
model: 'models/yolo/yolov12n.onnx',
classes: ['person', 'car'], // Filter classes or null for all
});
await detector.init();
const objects = await detector.detectFromCanvas(canvas);
```
### PoseDetector
```typescript
import { PoseDetector } from 'rtmlib-ts';
const detector = new PoseDetector({
detModel: 'models/yolo/yolov12n.onnx',
poseModel: 'models/rtmpose/end2end.onnx',
});
await detector.init();
const people = await detector.detectFromCanvas(canvas);
```
## Model Paths (Relative)
When using the library, reference models with relative paths from your web root:
```typescript
// From web root (rtmlib-ts/)
const detector = new ObjectDetector({
model: './models/yolo/yolov12n.onnx',
});
// From examples/
const detector = new ObjectDetector({
model: '../models/yolo/yolov12n.onnx',
});
```
## Performance
### YOLO12n (Object Detection)
| Backend | Input Size | Inference Time |
|---------|------------|----------------|
| WASM | 640Γ640 | ~80ms |
| WASM | 416Γ416 | ~40ms |
| WebGPU | 640Γ640 | ~30ms |
### RTMW (Pose Estimation)
| Backend | Input Size | Inference Time (per person) |
|---------|------------|---------------------------|
| WASM | 384Γ288 | ~25ms |
| WebGPU | 384Γ288 | ~10ms |
## COCO Classes (80)
YOLO12n detects all 80 COCO classes:
```
0: person 20: elephant 40: cup 60: toilet
1: bicycle 21: bear 41: fork 61: tv
2: car 22: zebra 42: knife 62: laptop
3: motorcycle 23: giraffe 43: spoon 63: mouse
4: airplane 24: backpack 44: bowl 64: remote
5: bus 25: umbrella 45: banana 65: keyboard
6: train 26: handbag 46: apple 66: cell phone
7: truck 27: tie 47: sandwich 67: microwave
8: boat 28: suitcase 48: orange 68: oven
9: traffic light 29: frisbee 49: broccoli 69: toaster
10: fire hydrant 30: skis 50: carrot 70: sink
11: stop sign 31: snowboard 51: hot dog 71: refrigerator
12: parking meter 32: sports ball 52: pizza 72: book
13: bench 33: kite 53: donut 73: clock
14: bird 34: baseball bat 54: cake 74: vase
15: cat 35: baseball glove 55: chair 75: scissors
16: dog 36: skateboard 56: couch 76: teddy bear
17: horse 37: surfboard 57: potted plant 77: hair drier
18: sheep 38: tennis racket 58: bed 78: toothbrush
19: cow 39: bottle 59: dining table
```
## License
- YOLO12: AGPL-3.0 (Ultralytics)
- RTMW: Apache-2.0 (OpenMMLab)
|