Download README.md from EnYa32/LargeFishClassifcationResNet50: direct link, hf CLI and curl.
- Browser
- Download file 2.81 kB
-
https://huggingface.co/spaces/EnYa32/LargeFishClassifcationResNet50/resolve/main/README.md
- Command line
-
hf download hf://spaces/EnYa32/LargeFishClassifcationResNet50/README.md
-
curl -L -o README.md https://huggingface.co/spaces/EnYa32/LargeFishClassifcationResNet50/resolve/main/README.md
A newer version of the Streamlit SDK is available: 1.67.0
title: FishClassifier_CNN_ResNet50
emoji: π
colorFrom: blue
colorTo: green
sdk: streamlit
app_file: src/streamlit_app.py
pinned: false
short_description: CNN with ResNet50 for 9-class fish classification.
license: mit
π Fish Species Classification β ResNet50 Transfer Learning
This project uses Transfer Learning with ResNet50 to classify images of 9 different fish species.
The model was trained using the A Large Scale Fish Dataset from Kaggle.
The goal of this project is to build a strong image classifier as part of my Deep Learning assignments.
I also built a CNN model without transfer learning for comparison.
π¦ Dataset
The dataset contains 9 fish classes:
- Black Sea Sprat
- Gilt-Head Bream
- Horse Mackerel
- Red Mullet
- Red Sea Bream
- Sea Bass
- Shrimp
- Striped Red Mullet
- Trout
The dataset is balanced and includes several thousand images.
Source: https://www.kaggle.com/datasets/crowww/a-large-scale-fish-dataset
π§ Model Overview (ResNet50)
This project uses ResNet50 with ImageNet weights as a feature extractor.
Model pipeline:
- Load ResNet50 without top layers (
include_top=False) - Freeze all base layers
- Add custom layers:
- GlobalAveragePooling2D
- Dense(256, activation='relu')
- Dropout(0.5)
- Dense(9, activation='softmax')
Training Settings
- Loss:
sparse_categorical_crossentropy - Optimizer: Adam (learning rate reduced automatically)
- Metrics:
- Accuracy
- Top-3 Accuracy
- Callbacks:
- EarlyStopping
- ReduceLROnPlateau
- ModelCheckpoint (saves the best model)
π Model Performance
Results on validation data:
- Validation Accuracy: ~98.5%
- Top-3 Accuracy: 100%
- Validation Loss: ~0.05
The ResNet50 model performs extremely well and clearly outperforms the CNN built from scratch.
π How to Use the Model
Load model:
import tensorflow as tf
model = tf.keras.models.load_model("best_fish_resnet50.keras")
Predict:
python
import numpy as np
from tensorflow.keras.preprocessing import image
img = image.load_img("your_image.jpg", target_size=(224, 224))
img_array = np.expand_dims(image.img_to_array(img), axis=0)
pred = model.predict(img_array)
print(pred)
π Streamlit App
This HuggingFace Space includes an interactive Streamlit app where you can:
Upload a fish image
See the model prediction
View confidence scores for all 9 classes
π Project Files
File Description
best_fish_resnet50.keras Saved ResNet50 model
streamlit_app.py Streamlit app code
requirements.txt Required Python packages
ResNet50_Fish_Report.pdf Technical report
fish_resnet50_history.pkl Training history
## π€ Author
**Enes Yamac**
Machine Learning & Computer Vision Project (2025)