Top AI Repos — open-source AI, indexed and scored

Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.

leondgarse/Keras_insightface

Insightface Keras implementation

44MITPython

View on GitHub

Stars: 263
Forks: 59
Watchers: 5
Open issues: 2
Contributors: 2

Star history, last 90 days— over 5 snapshots

2026-07-24 · 2632026-08-06 · 263

Star history, last 90 days
Date	Stars
2026-07-24	263
2026-07-25	263
2026-07-28	263
2026-07-30	263
2026-08-06	263

Today

— stars today

This week

— stars this week

This month

— stars this month

Momentum

0.0

growth rate 0.00%/day

From the README

# ___Keras insightface___
***

# License
  This is the keras implementation of [deepinsight/insightface](https://github.com/deepinsight/insightface), and is released under the MIT License. There is no limitation for both academic and commercial usage.

  The training data containing the annotation (and the models trained with these data) are available for non-commercial research purposes only.
# Table of Contents
<!-- TOC depthFrom:1 depthTo:6 withLinks:1 updateOnSave:1 orderedList:0 -->
- [Current accuracy](#current-accuracy)
- [Usage](#usage)
  - [Environment](#environment)
  - [Beforehand Data Prepare](#beforehand-data-prepare)
  - [Project Structure](#project-structure)
  - [Basic Training](#basic-training)
  - [Other Basic Functions and Parameters](#other-basic-functions-and-parameters)
  - [Learning rate](#learning-rate)
  - [Mixed precision float16](#mixed-precision-float16)
  - [Optimizers](#optimizers)
  - [Multi GPU train using horovod or distribute strategy](#multi-gpu-train-using-horovod-or-distribute-strategy)
- [Sub Center ArcFace](#sub-center-arcface)
- [Knowledge distillation](#knowledge-distillation)
- [Evaluating on IJB datasets](#evaluating-on-ijb-datasets)
- [TFLite model inference time test on ARM64](#tflite-model-inference-time-test-on-arm64)
- [Related Projects](#related-projects)
- [Citing](#citing)
<!-- /TOC -->
***

# Current accuracy
  - [Some comparing on EfficientNetV2_b0 with activation / data augmentation / loss function / others](https://github.com/leondgarse/Keras_insightface/discussions/42)
  - Model structures may change due to changing default behavior of building models.
  - `IJBB` and `IJBC` are scored at `TAR@FAR=1e-4`
  - Links in `Model backbone` are `h5` models in Google drive. Links in `Training` are training details.
  - The `r18` / `r34` / `r50` / `r100` on `glint360k` are models loaded weights from official publication.
  - `r50 magface` and `r100 magface` are ported from [Github IrvingMeng/MagFace](https://github.com/IrvingMeng/MagFace).
  - `r100 4m adaface` and `r100 12m adaface` are ported from [Github mk-minchul/AdaFace](https://github.com/mk-minchul/AdaFace).
  - **Please note `WebFace4M` / `WebFace12M` pretrained models cannot be used for any commercial purposes: [WebFace](https://www.face-benchmark.org/download.html).**

  | Model backbone | Training | lfw      | cfp_fp   | agedb_30 | IJBB     | IJBC     |
  | -------------- | ----- | -------- | -------- | -------- | -------- | -------- |
  | [Resnet34](https://github.com/leondgarse/Keras_insightface/releases/download/v1.0.0/resnet34_MXNET_E_SGD_REG_1e3_on_batch_true_lr1e1_random0_arc_S32_E1_BS512_casia_basic_agedb_30_epoch_36_0.949500.h5) | [CASIA, E40](https://github.com/leondgarse/Keras_insightface/discussions/36)  | 0.994667 | 0.949143 |   0.9495 |          |          |
  | [Mobilenet emb256](https://drive.google.com/file/d/1i0B6Hy1clGgfeOYtUXVPNveDEe2DTIBa/view?usp=sharing) | [Emore,E110](https://github.com/leondgarse/Keras_insightface/discussions/15#discussioncomment-286398) | 0.996000 | 0.951714 | 0.959333 | 0.887147 | 0.911745 |
  | [Mobilenet distill](https://drive.google.com/file/d/1yUjCG5rMeVCKTSPbST2F9BrRRlkDPzEA/view?usp=sharing) | [MS1MV3,E50](https://github.com/leondgarse/Keras_insightface/discussions/30) | 0.997333    | 0.969    | 0.975333 | 0.91889   | 0.940328 |
  | [se_mobile_facenet](https://github.com/leondgarse/Keras_insightface/releases/download/v1.0.0/TF11_se_mobilefacenet_pointwise_GDC_arc_emb256_dr0_sgd_no_alpha_bs512_ms1m_rand_0_bnm09_bne1e4_cos16_batch_float16_basic_agedb_30_epoch_50_0.973000.h5) | [MS1MV3,E50](https://github.com/leondgarse/Keras_insightface/discussions/15#discussioncomment-5904827) | 0.997333 | 0.969286 | 0.973000 | 0.922103 | **0.941913** |
  | [Ghostnet,S2,swish](https://github.com/leondgarse/Keras_insightface/releases/download/v1.0.0/ghostnet_130_960_s2_swish_se_swish_imagenet_bs512_lr01_test_basic_agedb_30_epoch_48_0.973667.h5) | [MS1MV3,E50](https://github.com/leondgarse/Ke

Excerpt of 52,161 characters

Read on GitHub

Top contributors

No contributor data yet — it is fetched during the contributor sync pass.

Quality score

Would you bet a product on this? Bounded 0–100 and slow moving.

44/10044

Maintenance: 5%; How recently the default branch was pushed to
Releases: 10%; Recency and cadence of tagged releases
Community: 9%; Contributor breadth and bus factor
Issue backlog: 100%; Open issues measured against the size of the audience
Documentation: 100%; README depth, plus a homepage and topics
Licensing: 100%; How freely the code can be adopted

What to know before adopting

No commits in over a year
No release in over a year
Bus factor of 1 — one contributor dominates
Usable in commercial products with attribution.

Details

Created: 7 years ago
Last push: 2 years ago
Last release: 5 years ago
Release tag: v1.0.0
Releases in the last year: 0
Default branch: master
Size: 48k KB
Owner type: User
Owner location: —
Discovered via: topic:tensorflow
First indexed: 13 days ago
Last synced: 12 days ago

leondgarse/Keras_insightface

Insightface Keras implementation

44MITPython

View on GitHub

Stars: 263
Forks: 59
Watchers: 5
Open issues: 2
Contributors: 2

Star history, last 90 days— over 5 snapshots

2026-07-24 · 2632026-08-06 · 263

Star history, last 90 days
Date	Stars
2026-07-24	263
2026-07-25	263
2026-07-28	263
2026-07-30	263
2026-08-06	263

Today

— stars today

This week

— stars this week

This month

— stars this month

Momentum

0.0

growth rate 0.00%/day

From the README

# ___Keras insightface___
***

# License
  This is the keras implementation of [deepinsight/insightface](https://github.com/deepinsight/insightface), and is released under the MIT License. There is no limitation for both academic and commercial usage.

  The training data containing the annotation (and the models trained with these data) are available for non-commercial research purposes only.
# Table of Contents
<!-- TOC depthFrom:1 depthTo:6 withLinks:1 updateOnSave:1 orderedList:0 -->
- [Current accuracy](#current-accuracy)
- [Usage](#usage)
  - [Environment](#environment)
  - [Beforehand Data Prepare](#beforehand-data-prepare)
  - [Project Structure](#project-structure)
  - [Basic Training](#basic-training)
  - [Other Basic Functions and Parameters](#other-basic-functions-and-parameters)
  - [Learning rate](#learning-rate)
  - [Mixed precision float16](#mixed-precision-float16)
  - [Optimizers](#optimizers)
  - [Multi GPU train using horovod or distribute strategy](#multi-gpu-train-using-horovod-or-distribute-strategy)
- [Sub Center ArcFace](#sub-center-arcface)
- [Knowledge distillation](#knowledge-distillation)
- [Evaluating on IJB datasets](#evaluating-on-ijb-datasets)
- [TFLite model inference time test on ARM64](#tflite-model-inference-time-test-on-arm64)
- [Related Projects](#related-projects)
- [Citing](#citing)
<!-- /TOC -->
***

# Current accuracy
  - [Some comparing on EfficientNetV2_b0 with activation / data augmentation / loss function / others](https://github.com/leondgarse/Keras_insightface/discussions/42)
  - Model structures may change due to changing default behavior of building models.
  - `IJBB` and `IJBC` are scored at `TAR@FAR=1e-4`
  - Links in `Model backbone` are `h5` models in Google drive. Links in `Training` are training details.
  - The `r18` / `r34` / `r50` / `r100` on `glint360k` are models loaded weights from official publication.
  - `r50 magface` and `r100 magface` are ported from [Github IrvingMeng/MagFace](https://github.com/IrvingMeng/MagFace).
  - `r100 4m adaface` and `r100 12m adaface` are ported from [Github mk-minchul/AdaFace](https://github.com/mk-minchul/AdaFace).
  - **Please note `WebFace4M` / `WebFace12M` pretrained models cannot be used for any commercial purposes: [WebFace](https://www.face-benchmark.org/download.html).**

  | Model backbone | Training | lfw      | cfp_fp   | agedb_30 | IJBB     | IJBC     |
  | -------------- | ----- | -------- | -------- | -------- | -------- | -------- |
  | [Resnet34](https://github.com/leondgarse/Keras_insightface/releases/download/v1.0.0/resnet34_MXNET_E_SGD_REG_1e3_on_batch_true_lr1e1_random0_arc_S32_E1_BS512_casia_basic_agedb_30_epoch_36_0.949500.h5) | [CASIA, E40](https://github.com/leondgarse/Keras_insightface/discussions/36)  | 0.994667 | 0.949143 |   0.9495 |          |          |
  | [Mobilenet emb256](https://drive.google.com/file/d/1i0B6Hy1clGgfeOYtUXVPNveDEe2DTIBa/view?usp=sharing) | [Emore,E110](https://github.com/leondgarse/Keras_insightface/discussions/15#discussioncomment-286398) | 0.996000 | 0.951714 | 0.959333 | 0.887147 | 0.911745 |
  | [Mobilenet distill](https://drive.google.com/file/d/1yUjCG5rMeVCKTSPbST2F9BrRRlkDPzEA/view?usp=sharing) | [MS1MV3,E50](https://github.com/leondgarse/Keras_insightface/discussions/30) | 0.997333    | 0.969    | 0.975333 | 0.91889   | 0.940328 |
  | [se_mobile_facenet](https://github.com/leondgarse/Keras_insightface/releases/download/v1.0.0/TF11_se_mobilefacenet_pointwise_GDC_arc_emb256_dr0_sgd_no_alpha_bs512_ms1m_rand_0_bnm09_bne1e4_cos16_batch_float16_basic_agedb_30_epoch_50_0.973000.h5) | [MS1MV3,E50](https://github.com/leondgarse/Keras_insightface/discussions/15#discussioncomment-5904827) | 0.997333 | 0.969286 | 0.973000 | 0.922103 | **0.941913** |
  | [Ghostnet,S2,swish](https://github.com/leondgarse/Keras_insightface/releases/download/v1.0.0/ghostnet_130_960_s2_swish_se_swish_imagenet_bs512_lr01_test_basic_agedb_30_epoch_48_0.973667.h5) | [MS1MV3,E50](https://github.com/leondgarse/Ke

Excerpt of 52,161 characters

Read on GitHub

Top contributors

No contributor data yet — it is fetched during the contributor sync pass.

Quality score

Would you bet a product on this? Bounded 0–100 and slow moving.

44/10044

Maintenance: 5%; How recently the default branch was pushed to
Releases: 10%; Recency and cadence of tagged releases
Community: 9%; Contributor breadth and bus factor
Issue backlog: 100%; Open issues measured against the size of the audience
Documentation: 100%; README depth, plus a homepage and topics
Licensing: 100%; How freely the code can be adopted

What to know before adopting

No commits in over a year
No release in over a year
Bus factor of 1 — one contributor dominates
Usable in commercial products with attribution.

Details

Created: 7 years ago
Last push: 2 years ago
Last release: 5 years ago
Release tag: v1.0.0
Releases in the last year: 0
Default branch: master
Size: 48k KB
Owner type: User
Owner location: —
Discovered via: topic:tensorflow
First indexed: 13 days ago
Last synced: 12 days ago

Top AI Repos — open-source AI, indexed and scored

leondgarse/Keras_insightface

From the README

Top contributors

Quality score

What to know before adopting

Categories

Details

leondgarse/Keras_insightface

From the README

Top contributors

Quality score

What to know before adopting

Categories

Details

Top AI Repos — open-source AI, indexed and scored

Trend

From the README

Top contributors

Quality score

What to know before adopting

Categories

Details

Trend

From the README

Top contributors

Quality score

What to know before adopting

Categories

Details