search

SthPhoenix / InsightFace-REST

InsightFace REST API for easy deployment of face recognition services with TensorRT in Docker.
Apache License 2.0
503 stars 117 forks source link

model failed to download #75

Open MyraBaba opened 2 years ago

MyraBaba commented 2 years ago

Hi,

when run insight_test.py it says :

RuntimeError: Failed downloading url http://insightface.ai/files/models/arcface_r100_v1.zip

I assume it ws downloading from google drive ? but here is not found

Best

PS: Consider to add Yolo5Face . Landrmarks could be better than scrfd

SthPhoenix commented 2 years ago

Hi! this is older debug script based on old versions of official insightface pip package, now it's not working because official package have changed and depreciated some models.

I'll look into yolo5 face detector, as I have said you before it seems promising.

MyraBaba commented 2 years ago

According to your best practices which recognition model and detection model is have better accuracy ?

Glintrx.onnx + scrfd10g ?

On 18 Mar 2022, at 13:31, SthPhoenix @.***> wrote:

Hi! this is older debug script based on old versions of official insightface pip package, now it's not working because official package have changed and depreciated some models.

I'll look into yolo5 face detector, as I have said you before it seems promising.

— Reply to this email directly, view it on GitHub https://github.com/SthPhoenix/InsightFace-REST/issues/75#issuecomment-1072279405, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEFRZH4X5NJL56BRD4UI573VARLR5ANCNFSM5RBNEKMA. Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub. You are receiving this because you authored the thread.

SthPhoenix commented 2 years ago

Hi! As for detection I would recommend using yolov5m or even yolov5l for best accuracy and recall including hard faces (strongly rotated ones etc.) But if your use case dont expect such faces scrfd_10g is a good fit. As for recognition I would use either glinr100 either w600k_r50 - latter is almost twice faster while preserve almost same accuracy, in some scenarios even better. I.e. w600k_r50 might give lower scores for same faces under very different conditions, but also it gives much lower scores for false positives, so in case you need to compare faces coming from same cameras in same conditions w600k_r50 might be a better option. In case of searching by image in database it might be a bit worse - it'll found person if it is in a DB, but might lower the score, which may harden automatic verification.