Update README.md
Browse files
README.md
CHANGED
@@ -3,6 +3,49 @@ tags:
|
|
3 |
- image-classification
|
4 |
- timm
|
5 |
library_name: timm
|
6 |
-
license:
|
|
|
|
|
|
|
7 |
---
|
8 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
- image-classification
|
4 |
- timm
|
5 |
library_name: timm
|
6 |
+
license: mit
|
7 |
+
datasets:
|
8 |
+
- cm93/eurosat
|
9 |
+
pipeline_tag: image-classification
|
10 |
---
|
11 |
+
## Model Details
|
12 |
+
|
13 |
+
**Model type:** Convolutional Neural Network (CNN)
|
14 |
+
|
15 |
+
**Finetuned from model :** ResNet18 (pre-trained on ImageNet-1k)
|
16 |
+
|
17 |
+
### Model Sources
|
18 |
+
**Repository:** https://github.com/chathumal93/EuroSat-RGB-Classifiers
|
19 |
+
|
20 |
+
## Training Details
|
21 |
+
|
22 |
+
### Training Data
|
23 |
+
|
24 |
+
The dataset comprises JPEG composite chips extracted from Sentinel-2 satellite imagery, representing the Red, Green, and Blue bands. It encompasses 27,000 labeled and geo-referenced images across 10 Land Use and Land Cover (LULC) classes
|
25 |
+
|
26 |
+
### Training Procedure
|
27 |
+
**Preprocessing:** Standard image preprocessing including resizing, center cropping, normalization, and data augmentation techniques [RandomHorizontalFlip and RandomVerticalFlip]
|
28 |
+
|
29 |
+
### Training Hyperparameters
|
30 |
+
- **Learning rate:** 3e-5
|
31 |
+
- **Batch size:** 64
|
32 |
+
- **Optimizer:** AdamW
|
33 |
+
- **Scheduler:** PolynomialLR
|
34 |
+
- **Loss:** CrossEntropyLoss
|
35 |
+
- **Betas**=(0.9, 0.999)
|
36 |
+
- **Weight_decay**=0.01
|
37 |
+
- **Epochs:** 20
|
38 |
+
|
39 |
+
|
40 |
+
## Evaluation
|
41 |
+
|
42 |
+
### Results
|
43 |
+
| Model | Phase | Avg Loss | Accuracy |
|
44 |
+
|:----------------------------:|:----------:|:--------:|:---------:|
|
45 |
+
| resnet18-eurosat | Train | 0.097586 | 97.01% |
|
46 |
+
| | Validation | 0.071375 | 97.70% |
|
47 |
+
| | Test | 0.068443 | 97.74% |
|
48 |
+
|
49 |
+
| Model | Accuracy | Precision | Recall | F1 |
|
50 |
+
|:----------------------------:|:--------:|:------------:|:-----------:|:-----------:|
|
51 |
+
| resnet18-eurosat | 97.74% | 0.97747 | 0.97741 | 0.97740 |
|