https://github.com/dennishnf/caffe-cnn-scripts
Scripts for iterative training with multiple models and datasets using Caffe framework.
https://github.com/dennishnf/caffe-cnn-scripts
artificial-intelligence caffe caffe-framework caffe-model caffe-prototxt deep-learning deep-learning-framework machine-learning
Last synced: about 1 year ago
JSON representation
Scripts for iterative training with multiple models and datasets using Caffe framework.
- Host: GitHub
- URL: https://github.com/dennishnf/caffe-cnn-scripts
- Owner: dennishnf
- License: gpl-3.0
- Created: 2020-05-23T03:24:06.000Z (almost 6 years ago)
- Default Branch: master
- Last Pushed: 2020-05-23T05:10:58.000Z (almost 6 years ago)
- Last Synced: 2025-01-23T10:36:53.145Z (about 1 year ago)
- Topics: artificial-intelligence, caffe, caffe-framework, caffe-model, caffe-prototxt, deep-learning, deep-learning-framework, machine-learning
- Language: Python
- Homepage:
- Size: 39.3 MB
- Stars: 2
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# caffe-cnn-scripts #
Scripts for iterative training with multiple models and datasets.
## a. for a single iteration ##
1. With scripts/generate_lmdb.py generate the "train.txt"/"test.txt", "train_lmdb"/"test_lmdb", "mean_image.binaryproto" files.
2. Verify models/model-NN/ model_train_val.prototxt model_deploy.prototxt the inputs, parameters of architecture like the size of conv, outputs.
3. Verify in models/model-NN/model_solver.prototxt the model and iterations.
4. Train the network, in terminal:
```
$ cd /home/dennis/Desktop/cnn-caffe-scripts/
```
```
$ caffe train --solver /home/dennis/Desktop/cnn-caffe-scripts/models/model-NN/model_solver.prototxt --gpu 0
```
Note: Fine tunning:
```
$ caffe train --solver /home/dennis/Desktop/cnn-caffe-scripts/models/model-NN/model_solver.prototxt --weights /home/dennis/Desktop/cnn-caffe-scripts/models/model-NN/bvlc_alexnet.caffemodel --gpu 0
```
Note: Resume training:
```
$ caffe train --solver /home/dennis/Desktop/cnn-caffe-scripts/models/model-NN/model_solver.prototxt --snapshot /home/dennis/Desktop/cnn-caffe-scripts/models/model-NN/train_iter_10000.solverstate
```
5. Test the network, in terminal:
```
$ cd /home/dennis/Desktop/cnn-caffe-scripts/
```
```
$ python /home/dennis/Desktop/cnn-caffe-scripts/scripts/testing_v2.py --proto /home/dennis/Desktop/cnn-caffe-scripts/models/model-NN/model_deploy.prototxt --model /home/dennis/Desktop/cnn-caffe-scripts/models/model-NN/train_iter_6000.caffemodel --mean /home/dennis/Desktop/cnn-caffe-scripts/input/dataset-MM/mean_image.binaryproto --txt /home/dennis/Desktop/cnn-caffe-scripts/input/dataset-MM/test.txt --cm none
```
```
$ python /home/dennis/Desktop/cnn-caffe-scripts/scripts/testing_v1.py --proto /home/dennis/Desktop/cnn-caffe-scripts/models/model-NN/model_deploy.prototxt --model /home/dennis/Desktop/cnn-caffe-scripts/models/model-NN/train_iter_6000.caffemodel --mean /home/dennis/Desktop/cnn-caffe-scripts/input/dataset-MM/mean_image.binaryproto --lmdb /home/dennis/Desktop/cnn-caffe-scripts/input/dataset-MM/test_lmdb
```
6. Draw the model and see the colvolutional filters:
```
cd /home/dennis/Desktop/cnn-caffe-scripts/scripts
```
```
$ python /home/dennis/caffe/python/draw_net.py /home/dennis/Desktop/cnn-caffe-scripts/models/model01/model_train_val.prototxt /home/dennis/Desktop/cnn-caffe-scripts/models/model01/model.png
```
```
$ python visualize_example.py
```
## b. for single and multiple iterations ##
1. With scripts/generate_lmdb.py generate the "train.txt"/"test.txt", "train_lmdb"/"test_lmdb", "mean_image.binaryproto" files.
2. Verify models/model-NN/ model_train_val.prototxt model_deploy.prototxt the inputs, parameters of architecture like the size of conv, outputs.
3. Verify in models/model-NN/model_solver.prototxt the model and iterations.
4. Then, use scripts/recursive_train.py, this script automatically train, test, draw models, and show convolutional filters.
## License ##
GNU General Public License, version 3 (GPLv3).
## About ##
By: Dennis Núñez-Fernández
Website: [http://dennishnf.com](http://dennishnf.com)