# Matlab codes for evaluation on MegaFace
Requirement:
1. [MatMTCNN](https://github.com/happynear/MTCNN_face_detection_alignment/tree/master/code/codes/vs/MatCascadeFaceDetection).
If you are not using Windows, you need to modify some of the codes to directly use the [Matlab version of MTCNN](https://github.com/kpzhang93/MTCNN_face_detection_alignment). If you have done such a work, I'm glad to merge your codes.
2. [Megaface](http://megaface.cs.washington.edu/participate/challenge.html). Please download
`Our dataset`, `FaceScrub full tgz`, `FaceScrub bounding boxes actors txt`,
`FaceScrub bounding boxes actresses txt`, `Linux Development Kit`.
Procedure:
1. Align Facescrub: `align_facescrub.m`. Then mannully confirm some failed samples through `align_facescrub_failures.m`.
2. Align Megaface: `align_megaface_from_list.m`. Then align failed samples through `align_megaface_failures.m`.
3. Extract Features using `extract_facescrub_feature.m`, `extract_megaface_feature.m`.
Alignment Logic:
The provided 3-point labels are the most accurate. The second accurate information is the bounding box. All 5-point labels are totally wrong.
So our logic is as follows.
1. `align_megaface_from_list.m`: If there is 3-point label, rotate and crop the image according to it. Then detect and align face from the cropped image.
2. `align_megaface_failures.m`: If there is no 3-point label or failed to detect from the cropped image, detect face from the raw image.
If the detected face and the given bounding box's IoU is over 30%, align this face.
If the IoU is below 30%, use the last two networks of MTCNN to forcely get the face score and 5 keypoints from the cropped image based on the given bounding box.
If the face score is above 0.3, use the detected 5 points to align the face.
If all methods are failed, directly crop the middle area of a face as the aligned face.
没有合适的资源?快使用搜索试试~ 我知道了~
一些处理人脸数据集的脚本。_MATLAB_Python_下载.zip
共63个文件
m:36个
txt:10个
py:10个
1.该资源内容由用户上传,如若侵权请联系客服进行举报
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
版权申诉
0 下载量 48 浏览量
2023-04-26
11:15:34
上传
评论
收藏 2.25MB ZIP 举报
温馨提示
一些处理人脸数据集的脚本。_MATLAB_Python_下载.zip
资源推荐
资源详情
资源评论
收起资源包目录
一些处理人脸数据集的脚本。_MATLAB_Python_下载.zip (63个子文件)
FaceDatasets-master
CFEE
create_list.m 341B
LICENSE 1KB
CASIA
translate_name_list.py 4KB
face_scrub_name.txt 1KB
lfw_log.txt 5KB
match_names.py 2KB
align_webface.py 7KB
webface_id_name_list.txt 231KB
testCascadeCNN.py 1KB
webface_facescrub_overlap.txt 336B
overlap_remove.py 2KB
webface_lfw_overlap.txt 128B
lfw-names.txt 93KB
webface_lfw_overlap_detail.txt 793B
Pain
store2hdf5.m 3KB
create_list.m 1KB
create_database.m 4KB
create_hdf5.m 2KB
analyse_dataset.m 184B
general_align.m 4KB
Megaface
extract_megaface_feature.m 3KB
face_deploy_20.prototxt 12KB
draw_curve.m 1KB
align_facescrub.m 7KB
parse_json.m 6KB
align_facescrub_failures.m 10KB
align_megaface_failures.m 6KB
ReadMe.md 2KB
extract_facescrub_feature.m 3KB
align_megaface_from_list.m 6KB
overview.m 5KB
KDEF
align_KDEF.m 3KB
create_list.m 630B
YTF
image_list.mat 3.1MB
extract_feature.m 1KB
roc_curve.m 661B
general_align.m 4KB
test_label_distribution.m 2KB
compare_feature.m 3KB
sample_feature.m 1KB
create_folder_map.m 695B
get_image_list_in_folder.m 855B
MS-Celeb
extract_and_align_ms_face.m 4KB
copy_cleaned_images.m 2KB
merge_CASIA_MS.m 1KB
extract_ms_face.m 1KB
base64decode.m 532B
get_image_list_in_folder.m 855B
.gitignore 1KB
CK
align_CK.m 3KB
create_list.m 358B
align_CK.py 7KB
VGGFace2
translate_name_list.py 4KB
vggface2_identity_trans.csv 314KB
convert_to_standard_list.py 601B
vggface2_facescrub_overlap.txt 144B
vggface2_lfw_overlap.txt 5KB
match_names.py 2KB
overlap_remove.py 2KB
LFW
pairs.txt 152KB
zzzgetPairs.m 2KB
getlfwPairs.m 1KB
README.md 1KB
共 63 条
- 1
资源评论
快撑死的鱼
- 粉丝: 1w+
- 资源: 9154
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功