- !69 Make all the transformers work with lists by default
- !47 Accept an annotator in FaceCrop: related to #26
- !70 Renamed the resources to use dash instead of under score: Renamed the resources to use dash instead of under score what we've discussed yesterday @ydayer
!72 Fix transformers naming in tests: Fixed some tests that failed after !70. Added temporary models
datafolder in .gitignore.
- !71 Face crop improvements
- !64 Dask pipelines
!75 [Preprocessor] Fix color channel: This fixes bugs introduced by commit 38e6503f. - Removing ambiguity in Base between
color_channelmethod (-> changed to
change_color_channelmethod) - Patching tests to send a list of arrays to Base.transform, not an array (as required)
- !76 Fixed Scaler and preprocessors that use annotations
- !77 TF2 pipelines: Hi @lcolbois, I'm opening an MR with your port. Let's do it one by one. Thanks for starting this work
- !78 Remove bob.db.msumfsd_mod
!79 Arcface from InsightFace: Hey @amohammadi, Shall I add
!73 Adaptation of Annotator to Tranformer: Removes facedetect and facelandmark. MTCNN follows the Transformer interface. The implementation of MTCNN
annotations()(in bob.ip.tensoflow_extractor) must be changed to accept a batch of images.
- !82 Cleanup: Making builds green again
memory_demandingfor TF based transformers: There are some situations where you have
SampleSetswith more than 1000 samples and we can't just TF
forwardthem in one shot without OOM. In this MR I introduce the argument
memory_demanding, where, once it's set to true, will `forward one sample at a time.
- !80 Port display-face-annotations script: Created a click command for the display-face-annotations script and fixed tests.
- !84 Test Facenet: This completes the testing of FaceNet by comparing with the embeddings generated in the TF1 version. N.B : Currently the test does NOT succeed bob/bob.bio.face!
- !81 Sample Loaders able to handle certain type of annotations...: .... and the implementation of the MEDS dataset as file lists
- !86 Some fixes to allow vulnerability tests with replay-attack and replay-mobile database
- !87 Renamed CSVDatasetDevEval to CSVDataset
- !89 Updates: - Added MORPH dataset - Proper documented the MEDS database - Fixed eyes annotation issue
- !88 SampleLoader as transformer: Depends on bob.pipelines!60 ping @ydayer @amohammadi
- !90 Updated Capeal database
!92 MultiFaceCrop: Test and describe the MultiFaceCrop preprocessor Integrates it into the baselines Now can run instructions such as
bob bio pipelines vanilla-biometrics multipie-pose inception-resnetv2-msceleband have the MultiFaceCrop automatically selected.
- !93 Small fixes the database tests so it flawlessly passes in [mac] builds
!94 Fixing the temp directory path for legacy algorithms IF checkpointing is off: Hi @lcolbois, Thanks for detecting the problem and opening the issue with all the details. The problem is that most of the legacy algorithms are not pickled serializable, therefore, we can't wrap their transformations as :any:
bob.pipelines.Sampleand use pipelines in memory (
--checkpointoption). Hence, we need to make them automatically "checkpointable" and use their legacy functions to persist data into the disk in a temporary directory. This was wrongly set in the
/tmp/was being set by default and this is not valid while running stuff on cluster. Basically, you are persisting data on the
/tmpof every node and obviously, this directory is not visible in other nodes. That's why things work locally and on the grid with checkpoints. In this MR bob.bio.base!238 I formalized a function that generates proper temporary directories that are visible on the cluster. closes #39
!91 Porting databases to the CSV interface: Hey, I'm starting from Mobio and will probably do MultiPIE (ping @lcolbois) tonight. Shall we deprecate the corresponding
bob.dbs? I would keep them around for a while...
- !95 Multipie Leaderboard: This introduces + Documentation on Multipie + Leaderboard with baseline results + A script for specifically plotting HTER in function of camera angle, with results
- !97 [Fix] Legacy helpers with fixed annotations: This fixes the helpers used for defining legacy baselines, when using fixed annotations. Beforehand, the instantiated cropper was still expecting to receive an "annotation" metadata, which broke the execution. I fixed a similar issue a while ago for Tensorflow baselines but missed to also fix legacy baselines. ping @tiago.pereira
- !99 [Fix][FaceCrop] Feed annotator with list of images: This fixes #40.
- !100 Resolve "EyesAnnotations ordering of x & y coordinates is wrong": Closes #41
- !96 TF2 training: Wrapping up an old script that trains some angular FR models (ArcFace, SphereFace, Modsoftmax) using different backbones
!101 Changed 112x112 crop: Changed the 112x112. In that way, the
arcface-insightfacegives ~0% on the Mobio dataset
[Min. criterion: EER ] Threshold on Development set `/idiap/temp/tpereira/temp/insight-face/scores-dev`: -2.459333e+01 ===================== ================ ================= .. Development Evaluation ===================== ================ ================= Failure to Acquire 0.00% 0.00% False Match Rate 0.03% (20/57960) 0.04% (65/147630) False Non Match Rate 0.04% (1/2520) 0.00% (0/3990) False Accept Rate 0.03% 0.04% False Reject Rate 0.04% 0.00% Half Total Error Rate 0.04% 0.02% ===================== ================ =================
!102 New baselines: Included two new baselines trained with ArcFace head These two were trained by myself using MSCeleb.
===================== ================ ================= .. Development Evaluation ===================== ================ ================= Failure to Acquire 0.0% 0.0% False Match Rate 0.5% (299/57960) 0.4% (531/147630) False Non Match Rate 0.5% (13/2520) 1.5% (59/3990) False Accept Rate 0.5% 0.4% False Reject Rate 0.5% 1.5% Half Total Error Rate 0.5% 0.9% ===================== ================ =================mobilenetv2-msceleb-arcface-2021
[Min. criterion: EER ] Threshold on Development set `/idiap/temp/tpereira/temp/mobilenetv2-mobio-male/scores-dev`: -9.859437e-01 ===================== ================ ================== .. Development Evaluation ===================== ================ ================== Failure to Acquire 0.0% 0.0% False Match Rate 1.1% (621/57960) 1.0% (1534/147630) False Non Match Rate 1.1% (27/2520) 2.5% (100/3990) False Accept Rate 1.1% 1.0% False Reject Rate 1.1% 2.5% Half Total Error Rate 1.1% 1.8% ===================== ================ ==================To use it do in a FR experiment do:
bob bio pipelines vanilla-biometrics <database> resnet50-msceleb-arcface-2021 ....
!103 Fix LFW database: - Ported some databases to the the CSVSampleLoaderBiometrics - Set the flag
allow_scoring_with_all_biometric_references=Falsein the LFW database entry-points. ping @mguenther
- !104 Resolve "ArcFace (MXNET) + IJBC lead to memory error": Closes #42
!105 New baseline: New baseline. Resnet50 (the one from keras) trained with VGG2 and arcface loss. Resource
- !107 [tests] Mark tensorflow tests as slow: related to #43
- !108 Replace nose with pytest: Related to bob#267
!109 Remove duplicated tests: Improved the
test_baselinetests to be tight with the score values, so we indirectly test the embeddings. Hence, we pretty much have the same coverage. The coverage decreased 1% due to some resources that are not activated, but it's fine. ~26 minutes and 17 GB ram usage much improved compared to 56 min and 28 GB RAM usage
- !110 Casia Africa Database: Created Casia-Africa database interface. - [x] Documentation - [x] Tests - [x] Resources
- !111 Integrating HTFace databases: Hi, I quickly integrated some HTFace databases. Here I ported the polathermal (VIS-Thermal) dataset and CBSR dataset (VIS-NIR) ping @amohammadi @ageorge @sebastien.marcel
- !60 Removed outdated COPYING and updated MANIFEST.in: This merge request fixes an inconsistency in the licenses. The copying file hadn't been removed and the MANIFEST.in file was still using it. This patch fixes both.
- !59 [mtcnn-annotator] replace it bob.ip.mtcnn with bob.ip.tensorflow_extractor
- !53 Do not allow upside down faces in FaceCrop by default: Fixes #33
- !54 Remove bob.ip.dlib annotator: bob.ip.dlib is no longer maintained.
- !52 Add Fargo database: New face recognition database.
- !55 added FARGO stuff wherever it was needed:
- !44 Fix the header level of index.rst
- !57 Improve the warning message in FaceCrop
- !56 Improve Replay Mobile interface
- Breaking changes:
- Dropped support for CSU baselines
- Detailed changes:
- !47 Accept an annotator in FaceCrop: related to #26
- !48 Dropped support to CSU baselines issue #29
- !51 Removing baselines: Related to this MR !49
- !50 Ijbc highlevel
!49 Refactoring baselines: Now we can all the facerec
baselines can be reached via
bob bio baselines --help. Refer to docs.
- Implemented IJBA customized plot
- Make the original_directory and annotation_directory a property
- Created IJBB high level interface
- Added script to display face annotations provided by a database.
- Migrate to conda based CI
- improve docs and tests
- !27 : Base preprocessor accept **kwargs
Outcome of the refactoring 2016 (new database interface).
42 commits in total
6ad4dade Increased latest version to 3.0.1b0 [skip ci]
e83032d3 Increased stable version to 3.0.0
80c8d778 Merge branch 'patch-doc' into 'master'
faf478f9 Add missing database import
ecda185b Merge branch 'patch-doc' into 'master'
fa7559c3 Point to gitlab pages
60de5d64 Merge branch 'replay' into 'master'
d09c7927 Add replay config and documentation
fa9a3e36 [ci] update the ci file
d3c29b15 Merge branch 'pre-release' into 'master'
15fc689b Point to gitlab.idiap.ch
cfc00481 Fixed sphinx warnings
1e36d500 Merge branch 'issue-8-remove-database-configuration' into 'master'
3afbe77f Merge branch 'issue-8-remove-database-configuration' into 'master'
d3859cf5 [refactoring2016] Added more tests
12bbf122 Removed temporary buildout
273a10a2 Merge branch 'issue-8-remove-database-configuration' into 'master'
a2460a0a [refactoring2016] Fixed small bug
9981d93d [refactoring2016] modifications in the contructor
4e2fbfe2 [refactoring2016] Removed the temporary modifications
dded0e08 [refactoring2016] Added ATNT entry point
1d83fbfd [refactoring2016] Fixed the extension
56f77097 Testing ci
d1e64fe8 Testing ci
94546899 Added bob.db.base in the eggs
c39ee0fd Added gridtk
5f6d61aa Temporary ci
4de8a012 [refactoring2016] Updated the annotation type
c117c223 [refactoring2016] Temporary ci file. This will not be merged
aa4903a7 [refactoring2016] Moved highlevel interface to bob.bio.face
52ced2e3 no need to test atnt here
a94ef624 Remove atnt config file since it is bob.bio.db
1024d805 [refactoring_2016] Removed references to bob.db.verification.utils
0b77c0fc [refactoring_2016] Removed bob.db.verification.utils
f3036b90 Removed dependency with bob.bio.base.DatabaseBob
49736214 Removed tests with FRGC
5809aba0 Added more packages
ae7fb2df [refactoring_2016] Migrating the database entry points to point to the High Level Interfaces
7b7b4a46 Increased latest version to 2.0.6b0 [skip ci]