esab commited on
Commit
12819b7
·
verified ·
1 Parent(s): 5db8beb

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +95 -0
README.md ADDED
@@ -0,0 +1,95 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # ResNet-18 Peripheral Blood Cell Classifier
2
+
3
+ ## Model Description
4
+
5
+ This is a ResNet-18 model fine-tuned for peripheral blood cell (PBC) classification using fastai. The model can classify blood cell images into 8 different cell types with 98.07% validation accuracy.
6
+
7
+ ## Model Details
8
+
9
+ - **Model Type**: ResNet-18 with transfer learning
10
+ - **Framework**: fastai (version <2.8.0)
11
+ - **Task**: Image Classification
12
+ - **Dataset**: Peripheral Blood Cell (PBC) dataset
13
+ - **Classes**: 8 cell types
14
+ - **Validation Accuracy**: 98.07%
15
+
16
+ ## Cell Types
17
+
18
+ The model can classify the following blood cell types:
19
+ 1. **Basophil** - A type of white blood cell involved in inflammatory reactions
20
+ 2. **Eosinophil** - White blood cells that fight parasites and allergic reactions
21
+ 3. **Erythroblast** - Immature red blood cells
22
+ 4. **IG (Immature Granulocyte)** - Immature white blood cells
23
+ 5. **Lymphocyte** - White blood cells that fight infections
24
+ 6. **Monocyte** - Large white blood cells that become macrophages
25
+ 7. **Neutrophil** - Most common white blood cells that fight bacterial infections
26
+ 8. **Platelet** - Cell fragments that help blood clotting
27
+
28
+ ## Training Details
29
+
30
+ - **Training Images**: 13,674
31
+ - **Validation Images**: 3,418
32
+ - **Architecture**: Pretrained ResNet-18 backbone with custom head
33
+ - **Training Strategy**:
34
+ - 4 epochs with frozen backbone
35
+ - 6 epochs with fine-tuning
36
+ - **Input Size**: 224x224 pixels
37
+ - **Preprocessing**: Standard ImageNet normalization
38
+
39
+ ## Performance
40
+
41
+ - **Validation Accuracy**: 98.07%
42
+ - **All cell types**: >95% precision and recall
43
+ - **Best performers**: Eosinophil and Platelet (100% precision)
44
+
45
+ ## Usage
46
+
47
+ ```python
48
+ from fastai.vision.all import *
49
+
50
+ # Load the model
51
+ learn = load_learner('cell_classifier.pkl')
52
+
53
+ # Predict on an image
54
+ pred, pred_idx, probs = learn.predict('path/to/blood_cell_image.jpg')
55
+ print(f"Predicted: {pred}")
56
+ print(f"Confidence: {probs[pred_idx]:.2%}")
57
+ ```
58
+
59
+ ## Requirements
60
+
61
+ ```
62
+ fastai>=2.7.0,<2.8.0
63
+ numpy<2.0
64
+ pillow>=10.0.0
65
+ ```
66
+
67
+ ## Model Files
68
+
69
+ - `cell_classifier.pkl` - Complete fastai learner with model and preprocessing
70
+ - `cell_classifier_weights.pth` - PyTorch weights only
71
+ - `confusion_matrix.png` - Validation confusion matrix
72
+ - `classification_report.csv` - Detailed classification metrics
73
+ - `training_summary.json` - Training configuration and results
74
+
75
+ ## Citation
76
+
77
+ If you use this model, please cite:
78
+
79
+ ```bibtex
80
+ @misc{pbc-cell-classifier-2024,
81
+ title={ResNet-18 Peripheral Blood Cell Classifier},
82
+ author={Your Name},
83
+ year={2024},
84
+ howpublished={Hugging Face Hub},
85
+ url={https://huggingface.co/your-username/pbc-cell-classifier}
86
+ }
87
+ ```
88
+
89
+ ## License
90
+
91
+ This model is released under the MIT License.
92
+
93
+ ## Created For
94
+
95
+ HuggingFace Agents-MCP-Hackathon Track 1 - MCP Tool/Server