File size: 1,080 Bytes
ee2378c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: apache-2.0
tags:
- computer-vision
- image-segmentation
- welding
- weld-inspection
- yolo
---

# WeldVision Ensemble

Four-model YOLO segmentation ensemble for weld-surface defect inspection.

## Models

- `weights/best.pt`
- `weights/best_v0.pt`
- `weights/crack_specialist.pt`
- `weights/spatters_specialist.pt`

Inference runs **locally** after downloading the weights from the Hugging Face
Hub. No Hugging Face Inference Endpoint is required.

## Install

```bash
pip install ultralytics huggingface_hub opencv-python-headless numpy
```

## Use

```python
from weldvision import WeldVision

model = WeldVision.from_pretrained(
    "bhavibhatt/weldvision-ensemble"
)

result = model.predict("weld.jpg")

print(result)
```

The first call downloads and caches the four weights. Subsequent calls reuse the
local Hugging Face cache.

## Important

This is a proof-of-concept weld surface inspection system. The score and
PASS/REVIEW/FAIL rules are prototype decision logic and are not welding-code
acceptance criteria or a substitute for qualified human inspection.