https://github.com/vfdev-5/caffe-tools
Tools to use with BVLC/Caffe
https://github.com/vfdev-5/caffe-tools
Last synced: about 1 year ago
JSON representation
Tools to use with BVLC/Caffe
- Host: GitHub
- URL: https://github.com/vfdev-5/caffe-tools
- Owner: vfdev-5
- Created: 2016-08-30T22:48:44.000Z (almost 10 years ago)
- Default Branch: master
- Last Pushed: 2016-10-03T15:35:59.000Z (over 9 years ago)
- Last Synced: 2025-02-08T10:43:56.612Z (over 1 year ago)
- Language: Python
- Homepage:
- Size: 1.35 MB
- Stars: 0
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# A set of tools to use with [BVLC/caffe](https://github.com/BVLC/caffe) for deep learning with images
All tools are written in python and inspired from ideas and algorithms found in various sources.
Tools are divided into categories:
- train/test data preprocessing
- training / validation / visualization
### Train/test data preprocessing
- ...
- ...
- ...
### Training / validation / visualization
- single image validation and features visualization for classification CNN
- `single_image_analyze.py` computes net forward pass on a single image and a 'heatmap' of trained features of the model detected on this image. See [here](https://github.com/vfdev-5/caffe-tools/blob/master/trainval/README.md) for more details.
For example, features 'heatmap' of the [famous cat](http://nbviewer.jupyter.org/github/BVLC/caffe/blob/master/examples/00-classification.ipynb) looks like
```
python trainval/single_image_analysis.py cat.jpg bvlc_reference_caffenet/deploy.prototxt bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel --mean ilsvrc12/imagenet_mean.binaryproto --verbose
```
- network model's graph visualization with matplotlib
- `draw_net.py` displays network model graph as figures using matplotlib. Make sure that you have *graphviz* and *pydotplus* installed.
```
python draw_net.py /path/to/model.prototxt relative/path/to/model2.prototxt
```
## Python 2.7 and dependencies :
* pycaffe from [BVLC/caffe](https://github.com/BVLC/caffe)
* PyYAML
* numpy
* matplotlib
* scipy
* sklearn
* pandas
* opencv
* lmdb
~~See `requirements.txt` file~~
## How to use
Setup your configuration in the `config.yaml`:
(you can rename and modify `config.yaml.example`)
```
# Path to Caffe repository
CAFFE_PATH: /path/to/caffe/
# Path to image dataset
DATASET_PATH: /Pascal_VOC_2011/VOCdevkit/VOC2011/JPEGImages/
# Path to labels file
LABELS_FILE_PATH: /path/to/data/ilsvrc12/synset_words.txt
# List of available nets: [ [, , , ], ... ]
NETS_LIST: {
'caffe_net': ['bvlc_reference_caffenet/deploy.prototxt', 'bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel', 'ilsvrc12/imagenet_mean.binaryproto'],
'vgg16': ['vgg/VGG_ILSVRC_16_layers_deploy.prototxt', 'vgg/VGG_ILSVRC_16_layers.caffemodel', 'ilsvrc12/imagenet_mean.binaryproto'],
'resnet50': ['ResNet/ResNet-50-deploy.prototxt', 'ResNet/ResNet-50-model.caffemodel', 'ilsvrc12/imagenet_mean.binaryproto']
}
# List with paths to models
MODELS_PATH_LIST: [
/path/to/caffe/models,
/path/to/fcn/voc-fcn16s
]
# List with paths to weights
RESOURCES_PATH_LIST: [
/path/to/models,
/path/to/fcn/voc-fcn16s
]
# Use CPU instead of GPU
USE_CPU: false
```