Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/yjg30737/pyqt-using-transfer-learning-resnet50-example
PyQt example of using transfer-learned ResNet50 model to distinguish 7 flowers
https://github.com/yjg30737/pyqt-using-transfer-learning-resnet50-example
image-classification image-classifier pyqt-ai pyqt5 resnet50 transfer-learning
Last synced: 7 days ago
JSON representation
PyQt example of using transfer-learned ResNet50 model to distinguish 7 flowers
- Host: GitHub
- URL: https://github.com/yjg30737/pyqt-using-transfer-learning-resnet50-example
- Owner: yjg30737
- License: mit
- Created: 2023-11-27T02:27:33.000Z (12 months ago)
- Default Branch: main
- Last Pushed: 2024-08-04T06:12:47.000Z (3 months ago)
- Last Synced: 2024-08-04T07:27:37.699Z (3 months ago)
- Topics: image-classification, image-classifier, pyqt-ai, pyqt5, resnet50, transfer-learning
- Language: Python
- Homepage:
- Size: 81.6 MB
- Stars: 2
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# pyqt-using-transfer-learning-resnet50-example
PyQt example of using transfer-learned ResNet50 model to distinguish 7 flowers
## What is ResNet50?
ResNet50 is an image classification model. While it is excellent on its own, it is often used in creating custom models due to its flexibility in transfer learning and fine-tuning.
## What is the difference between Transfer Learning and Fine Tuning
Note: This is personal opinion by my observation.
If you declare the transform to be applied to a transfer-learned model in a different way such as from
```python
transform = transforms.Compose([
transforms.Resize((img_height, img_width)),
transforms.CenterCrop(224),
transforms.Lambda(lambda img: img.convert('RGB')),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
```
to
```python
transform = transforms.Compose([
transforms.Resize((img_height, img_width)),
transforms.Lambda(lambda img: img.convert('RGB')),
transforms.ToTensor(),
])
```
the accuracy significantly decreases. For fine-tuned models, as long as the transform is appropriate for the model, the detailed structure does not impact the prediction accuracy.
You don't need code like this that explicitly prevents changes to the weights:
```python
for param in model.parameters():
param.requires_grad = False
```
Fine-tuning is a more advanced form of transfer learning. However, to understand the mechanism of fine-tuning, consider the following example.
For an example source of transfer learning a ResNet50 model to recognize seven types of flowers specified by the user, please refer to the following Kaggle notebook.
Also take a look at fine-tuning ResNet50 model example of PyQt: pyqt-using-finetuned-resnet50-example
## Requirements
* PyQt5 >= 5.14
* torch
* torchvision
* numpy
## How to Run
1. Do the transfer larning model with notebook in the link above, get result.pth and download it
2. git clone ~
3. Put the downloaded model in the root folder
4. pip install -r requirements.txt
5. python main.py
## Preview
