Update README.md
Browse files
README.md
CHANGED
@@ -1,5 +1,91 @@
|
|
1 |
-
|
2 |
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
|
5 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
|
3 |
+
pipeline_tag: image-classification
|
4 |
+
library_name: torchvision
|
5 |
+
tags:
|
6 |
+
- image-classification
|
7 |
+
- efficientnet
|
8 |
+
- efficientnet-v2
|
9 |
+
- garbage
|
10 |
+
- waste-sorting
|
11 |
+
metrics:
|
12 |
+
- accuracy
|
13 |
|
14 |
+
---
|
15 |
+
|
16 |
+
# Garbage Classifier – EfficientNet‑V2‑S
|
17 |
+
|
18 |
+
A finetuned EfficientNet‑V2‑S model that recognises **10 waste categories**
|
19 |
+
(battery, glass, plastic, etc.) for smart recycling and sorting applications.
|
20 |
+
|
21 |
+
| id | class |
|
22 |
+
| -: | ---------- |
|
23 |
+
| 0 | battery |
|
24 |
+
| 1 | biological |
|
25 |
+
| 2 | cardboard |
|
26 |
+
| 3 | clothes |
|
27 |
+
| 4 | glass |
|
28 |
+
| 5 | metal |
|
29 |
+
| 6 | paper |
|
30 |
+
| 7 | plastic |
|
31 |
+
| 8 | shoes |
|
32 |
+
| 9 | trash |
|
33 |
+
|
34 |
+
---
|
35 |
+
|
36 |
+
## Quick Start
|
37 |
+
|
38 |
+
```python
|
39 |
+
from huggingface_hub import InferenceClient
|
40 |
+
|
41 |
+
client = InferenceClient("attilaultzindur/garbage_classifier_effnetv2s_ft")
|
42 |
+
|
43 |
+
with open("your_image.jpg", "rb") as f:
|
44 |
+
predictions = client.post(data={"inputs": f.read()})
|
45 |
+
print(predictions)
|
46 |
+
```
|
47 |
+
|
48 |
+
Example output:
|
49 |
+
|
50 |
+
```json
|
51 |
+
[
|
52 |
+
{"label": "plastic", "score": 0.997},
|
53 |
+
{"label": "metal", "score": 0.002},
|
54 |
+
…
|
55 |
+
]
|
56 |
+
```
|
57 |
+
|
58 |
+
---
|
59 |
+
|
60 |
+
## Model Details
|
61 |
+
|
62 |
+
| Field | Value |
|
63 |
+
| ------------------------ | --------------------------------------------------------- |
|
64 |
+
| Architecture | EfficientNet‑V2‑S (torchvision) |
|
65 |
+
| Input size | `3 × 224 × 224` |
|
66 |
+
| Normalisation | mean = \[0.485 0.456 0.406], std = \[0.229 0.224 0.225] |
|
67 |
+
| Classification head | `Linear(1280→256) → ReLU → Dropout(0.5) → Linear(256→10)` |
|
68 |
+
| Best validation accuracy | **97.6 %** after 20 epochs |
|
69 |
+
|
70 |
+
### Training summary
|
71 |
+
|
72 |
+
* **Dataset:** [Garbage Classification v2 (Kaggle)](https://www.kaggle.com/datasets/sumn2u/garbage-classification-v2)
|
73 |
+
split 80 % train / 20 % val
|
74 |
+
* **Augmentations:** RandomResizedCrop, ColorJitter, RandomAffine, HorizontalFlip
|
75 |
+
* **Optimiser:** Adam, LR = 1 e‑4
|
76 |
+
* **Frozen layers:** first 70 % of feature blocks
|
77 |
+
* **Hardware:** single NVIDIA GPU
|
78 |
+
|
79 |
+
---
|
80 |
+
|
81 |
+
## Reproduce
|
82 |
+
|
83 |
+
The full training script is provided in `train_script.py`.
|
84 |
+
Run it with the same hyper‑parameters to reproduce the checkpoint.
|
85 |
+
|
86 |
+
---
|
87 |
+
|
88 |
+
## Licence
|
89 |
+
|
90 |
+
No licence has been specified yet.
|
91 |
+
Add an appropriate open‑source licence before using the model in production.
|