OrgLion-ML
/
openface
mirror of https://github.com/cmusatyalab/openface.git
1
0
Fork 0
Code Issues Releases Wiki Activity

Update docs with new align-dlib interface.

This commit is contained in:
Brandon Amos 2015-12-12 13:43:30 -05:00
parent b9c848c8a2
commit af575bdb72
4 changed files with 18 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
docs/accuracy.md
View File

@ -21,10 +21,10 @@ in `./data/lfw/raw` and `./data/lfw/deepfunneled`.
1. Install prerequisites as below.
2. Preprocess the raw `lfw` images, change `8` to however many
separate processes you want to run:
`for N in {1..8}; do ./util/align-dlib.py data/lfw/raw align affine data/lfw/dlib-affine-sz:96 --size 96 & done`.
`for N in {1..8}; do ./util/align-dlib.py data/lfw/raw align innerEyesAndBottomLip data/lfw/dlib-affine-sz:96 --size 96 & done`.
Fallback to deep funneled versions for images that dlib failed
to align:
`./util/align-dlib.py data/lfw/raw align affine data/lfw/dlib-affine-sz:96 --size 96 --fallbackLfw data/lfw/deepfunneled`
`./util/align-dlib.py data/lfw/raw align innerEyesAndBottomLip data/lfw/dlib-affine-sz:96 --size 96 --fallbackLfw data/lfw/deepfunneled`
3. Generate representations with `./batch-represent/main.lua -outDir evaluation/lfw.nn4.v1.reps -model models/openface/nn4.v1.t7 -data data/lfw/dlib-affine-sz:96`
4. Generate the ROC curve from the `evaluation` directory with `./lfw-roc.py --workDir lfw.nn4.v1.reps`.
This creates `roc.pdf` in the `lfw.nn4.v1.reps` directory.

15
docs/demo-3-classifier.md
View File

@ -41,7 +41,7 @@ person-m
### 2. Preprocess the raw images
Change `8` to however many
separate processes you want to run:
`for N in {1..8}; do ./util/align-dlib.py <path-to-raw-data> align affine <path-to-aligned-data> --size 96 & done`.
`for N in {1..8}; do ./util/align-dlib.py <path-to-raw-data> align innerEyesAndBottomLip <path-to-aligned-data> --size 96 & done`.
### 3. Generate Representations
`./batch-represent/main.lua -outDir <feature-directory> -data <path-to-aligned-data>`
@ -92,3 +92,16 @@ Run the classifier on your images with:
| Adams | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/adams.jpg' width='200px'></img> | AmyAdams | 0.98 |
| Lennon 1 (Unknown) | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/lennon-1.jpg' width='200px'></img> | DavidBoreanaz | 0.27 |
| Lennon 2 (Unknown) | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/lennon-2.jpg' width='200px'></img> | DavidBoreanaz | 0.43 |
# Minimal Working Example to Extract Features
```
openface(master*)$ mkdir -p classify-test/raw/{lennon,clapton}
openface(master*)$ cp images/examples/lennon-* classify-test/raw/lennon
openface(master*)$ cp images/examples/clapton-* classify-test/raw/clapton
openface(master*)$ ./util/align-dlib.py classify-test/raw align innerEyesAndBottomLip classify-test/aligned --size 96
openface(master*)$ ./batch-represent/main.lua -outDir classify-test/features -data classify-test/aligned
...
nImgs:  4
Represent: 4/4
```

2
docs/training-new-models.md
View File

@ -46,7 +46,7 @@ person-m
## 2. Preprocess the raw images
Change `8` to however many
separate processes you want to run:
`for N in {1..8}; do ./util/align-dlib.py <path-to-raw-data> align affine <path-to-aligned-data> --size 96 & done`.
`for N in {1..8}; do ./util/align-dlib.py <path-to-raw-data> align innerEyesAndBottomLip <path-to-aligned-data> --size 96 & done`.
Prune out directories with less than N (I use 10) images
per class with `./util/prune-dataset.py <path-to-aligned-data> --numImagesThreshold <N>` and
then split the dataset into `train` and `val` subdirectories

2
docs/visualizations.md
View File

@ -46,7 +46,7 @@ person-m
## 2. Preprocess the raw images
Change `8` to however many
separate processes you want to run:
`for N in {1..8}; do ./util/align-dlib.py <path-to-raw-data> align affine <path-to-aligned-data> --size 96 & done`.
`for N in {1..8}; do ./util/align-dlib.py <path-to-raw-data> align innerEyesAndBottomLip <path-to-aligned-data> --size 96 & done`.
## 3. Generate Representations
`./batch-represent/main.lua -outDir <feature-directory> -data <path-to-aligned-data>`