Rayzar142 commited on
Commit
4f2eb2c
·
verified ·
1 Parent(s): d80a2d4

Upload 3 files

Browse files
Files changed (3) hide show
  1. app.py +22 -0
  2. requirements.txt +7 -0
  3. sunfish_classifier.py +62 -0
app.py ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Gradio app for HuggingFace Spaces deployment
2
+ from fastai.vision.all import *
3
+ import gradio as gr
4
+ import pathlib
5
+
6
+ # Load model (ensure sunfish_model.pkl is in the same directory as this file)
7
+ learn = load_learner('sunfish_model.pkl')
8
+
9
+ def predict(img):
10
+ pred_class, pred_idx, outputs = learn.predict(img)
11
+ return {learn.dls.vocab[i]: float(outputs[i]) for i in range(len(outputs))}
12
+
13
+ # Create Gradio interface
14
+ demo = gr.Interface(
15
+ fn=predict,
16
+ inputs=gr.Image(type="pil"),
17
+ outputs=gr.Label(num_top_classes=3),
18
+ title="Sunfish Classifier",
19
+ description="Classify Lepomis (sunfish) species using fastai and ResNet18"
20
+ )
21
+
22
+ demo.launch()
requirements.txt ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ fastai==2.7.14
2
+ torch==2.2.2
3
+ torchvision==0.17.2
4
+ numpy==1.26.4
5
+ gradio
6
+ datasets
7
+ pandas
sunfish_classifier.py ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Sunfish Classifier Training Code (tweaked from cats vs dogs original)
2
+ # For Google Colab: run the !pip lines directly; for local Python, install dependencies via terminal first
3
+
4
+ # Install dependencies (Colab/Jupyter only - remove ! for local Python)
5
+ !pip install -Uqq fastai datasets pandas gradio
6
+
7
+ from fastai.vision.all import *
8
+ from datasets import load_dataset
9
+ import pandas as pd
10
+ from pathlib import Path
11
+ import os
12
+
13
+ # 1. Load Fish-Vista dataset (contains Lepomis/sunfish species)
14
+ print("Loading Fish-Vista dataset...")
15
+ ds = load_dataset("imageomics/fish-vista", name="species_classification", split="train")
16
+ ds = ds.with_format("image")
17
+
18
+ # 2. Filter for sunfish (Lepomis genus)
19
+ print("Filtering for Lepomis/sunfish species...")
20
+ lepomis_labels = [l for l in ds.features['label'].names if 'Lepomis' in l]
21
+ print(f"Found Lepomis labels: {lepomis_labels}")
22
+ lepomis_indices = [i for i, l in enumerate(ds['label']) if ds.features['label'].int2str(l) in lepomis_labels]
23
+ sunfish_ds = ds.select(lepomis_indices)
24
+ print(f"Total sunfish images: {len(sunfish_ds)}")
25
+
26
+ # 3. Save to local folder for fastai
27
+ path = Path('/tmp/sunfish_data')
28
+ if path.exists():
29
+ import shutil
30
+ shutil.rmtree(path)
31
+ path.mkdir(parents=True, exist_ok=True)
32
+ print(f"Saving images to {path}...")
33
+
34
+ labels = [sunfish_ds.features['label'].int2str(l) for l in sunfish_ds['label']]
35
+ image_paths = []
36
+ for i, (img, label) in enumerate(zip(sunfish_ds['image'], labels)):
37
+ # Clean label for folder name (replace spaces/special chars)
38
+ clean_label = label.replace(" ", "_").replace("/", "_")
39
+ class_dir = path / clean_label
40
+ class_dir.mkdir(exist_ok=True)
41
+ img_path = class_dir / f"{i}.jpg"
42
+ img.save(img_path)
43
+ image_paths.append((str(img_path), label))
44
+
45
+ # 4. Create fastai DataLoaders
46
+ print("Creating DataLoaders...")
47
+ df = pd.DataFrame(image_paths, columns=['path', 'label'])
48
+ dls = ImageDataLoaders.from_df(
49
+ df, path='/', valid_pct=0.2, seed=42,
50
+ fn_col='path', label_col='label', item_tfms=Resize(224)
51
+ )
52
+ print("Showing batch...")
53
+ dls.show_batch(max_n=9)
54
+
55
+ # 5. Train model
56
+ print("Training model...")
57
+ learn = vision_learner(dls, resnet18, metrics=error_rate)
58
+ learn.fine_tune(3)
59
+
60
+ # 6. Export model
61
+ learn.export('sunfish_model.pkl')
62
+ print("Model exported to sunfish_model.pkl")