pip install insightface and https://insightface.ai #891

nttstar · 2019-09-05T03:23:12Z

are both online.
Any suggestion is welcome!
For pip source code, please refer to https://github.com/deepinsight/insightface/tree/master/python-package

2021.05.15 Update:

pip insightface-0.2.0 is ready now. Please update with pip install -U insightface

For insightface pip-package <= 0.1.5, we use MXNet as inference backend, please download all models from onedrive, and put them all under ~/.insightface/models/ directory.

Starting from insightface>=0.2, we use onnxruntime as inference backend, please download our antelope model release from onedrive, and put it under ~/.insightface/models/, so there're onnx models at ~/.insightface/models/antelope/*.onnx.

The antelope model release contains ResNet100@Glint360K recognition model and SCRFD-10GF detection model. Please check deploy/test.py for detail.

The text was updated successfully, but these errors were encountered:

Talgin · 2019-09-05T06:19:43Z

Hi, @nttstar
I think you have an error on 47th line of face_recognition - you don't have model parameter in your class.

Regards,

nttstar · 2019-09-05T06:32:25Z

@Talgin We should call prepare() first then we will have self.model var.

Talgin · 2019-09-05T07:38:06Z

Oh, sorry, I see it now 👍

rebotnix · 2019-09-06T08:04:20Z

please add

from future import division
import mxnet as mx
import numpy as np
import mxnet.ndarray as nd
import cv2

in your face_detection.py

or you will get an issue when you change scaling call in your tutorial example when you want to resize the image cause cv2 is missing.

bboxes, landmark = model.detect(img, threshold=0.5, scale=0.5)

RomanSteinberg · 2019-09-09T09:03:58Z

Hi,

Relative imports is a bad style which can cause building errors. Consider

from ..model_zoo import model_zoo

replace with

from insightface.model_zoo import model_zoo

You should check PEP8.
There is no minimal example to run in README. It will also answer questions like @Talgin asked.
There is a good practice called RAII. Your prepare methods contradict this idiom. Please, consider refactoring.

RomanSteinberg · 2019-09-09T09:40:54Z

@nttstar I installed your package and ran the following code:

import cv2
import sys
import insightface

from PyQt5.QtWidgets import QApplication

# The following line changes results of gender-age model.
# gender == 1, if the following line is commented
# gender == 0, if the following line isn't commented
QApplication(sys.argv)  

fa = insightface.app.FaceAnalysis()
fa.prepare(-1)
img = cv2.imread('Tom_Hanks_54745.png')
face = fa.get(img)[0]
print(face.gender)

The comment in this code describes the problem I found. I created an issue #899 which can give you more information about this problem.

nttstar · 2019-09-09T09:50:38Z

@RomanSteinberg Quite strange. It should always output 1(male). What if put QApplication(sys.argv) this line in the end of your script?

RomanSteinberg · 2019-09-09T09:57:02Z

@nttstar it outputs 1

dexception · 2019-09-17T08:18:06Z

off topic:

@nttstar
Can you share some details on how to train RetinaNet on Resnet-18 or Resnet 10 to make it real time ?

khanhnt · 2019-10-10T07:59:55Z

Hi @nttstar,

Thank you for your work. While using FaceAnalysis, we met the issue where the program was unsuccessful in loading the model and fall in a loop with the following message:

src/nnvm/legacy_json_util.cc:209: Loading symbol saved by previous version v1.3.0. Attempting to upgrade...
src/nnvm/legacy_json_util.cc:217: Symbol successfully upgraded!
src/operator/nn/./cudnn/./cudnn_algoreg-inl.h:97: Running performance tests to find the best convolution algorithm, this can take a while... (set the environment variable MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 to disable)

We also got the same issue when using the detection (retina) and recogntion together, event the detection and the recogntion was executed on different GPU.
It's ok when runing two or more recogtion instances.
I tried to export MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 and also update cudnn_tune in R50-symbol.json to "off", but it is not ok.

Updated: the issues was fixed by increasing timeout value while loading model

ucasiggcas · 2019-11-27T11:28:55Z

hi,dear,
I want to use the mobilenet model,could it be used directly,for I do not want to install mxnet.
and if I install the insightface, must I install the mxnet ??

thx

Programmerwyl · 2019-12-19T03:10:13Z

Hi :
How to tell if two faces are the same person?
which parameter was used ？
embedding ，embedding_norm，normed_embedding

face = Face(bbox = bbox, landmark = landmark, det_score = det_score, embedding = embedding, gender = gender, age = age
                    , normed_embedding=normed_embedding, embedding_norm = embedding_norm)

Is the function ‘compute_sim’ used to calculate the distance between two faces?  how to set threshold

def compute_sim(self, img1, img2):

benouinirachid · 2020-01-27T09:23:05Z

Hi @nttstar,

Thank you for your work. While using FaceAnalysis, we met the issue where the program was unsuccessful in loading the model and fall in a loop with the following message:

src/nnvm/legacy_json_util.cc:209: Loading symbol saved by previous version v1.3.0. Attempting to upgrade...
src/nnvm/legacy_json_util.cc:217: Symbol successfully upgraded!
src/operator/nn/./cudnn/./cudnn_algoreg-inl.h:97: Running performance tests to find the best convolution algorithm, this can take a while... (set the environment variable MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 to disable)

We also got the same issue when using the detection (retina) and recogntion together, event the detection and the recogntion was executed on different GPU.
It's ok when runing two or more recogtion instances.
I tried to export MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 and also update cudnn_tune in R50-symbol.json to "off", but it is not ok.

Updated: the issues was fixed by increasing timeout value while loading model

Could you please provide more details on how you set the timeout ?

khanhnt · 2020-01-27T09:46:38Z

Hi @nttstar,
Thank you for your work. While using FaceAnalysis, we met the issue where the program was unsuccessful in loading the model and fall in a loop with the following message:

src/nnvm/legacy_json_util.cc:209: Loading symbol saved by previous version v1.3.0. Attempting to upgrade...
src/nnvm/legacy_json_util.cc:217: Symbol successfully upgraded!
src/operator/nn/./cudnn/./cudnn_algoreg-inl.h:97: Running performance tests to find the best convolution algorithm, this can take a while... (set the environment variable MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 to disable)

We also got the same issue when using the detection (retina) and recogntion together, event the detection and the recogntion was executed on different GPU.
It's ok when runing two or more recogtion instances.
I tried to export MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 and also update cudnn_tune in R50-symbol.json to "off", but it is not ok.
Updated: the issues was fixed by increasing timeout value while loading model

Could you please provide more details on how you set the timeout ?

We have deployed the model with Docker, so we just increased the timeout value of Docker.

xiashiwendao · 2020-05-04T12:45:26Z

In python2.7, pip install insightface, found it dependency networkx2.3, which need python3.5+, how to resolve it?

Error Stack:

... ....

Using cached https://mirrors.aliyun.com/pypi/packages/ea/0b/189cd3c19faf362ff2df5f301456c6cf8571ef6684644cfdfdbff293825c/cloudpickle-1.3.0-py2.py3-none-any.whl (26 kB)
Collecting networkx>=1.8
Using cached https://mirrors.aliyun.com/pypi/packages/bf/63/7b579dd3b1c49ce6b7fd8f6f864038f255201410905dd183cf7f4a3845cf/networkx-2.4.tar.gz (1.5 MB)
ERROR: Command errored out with exit status 1:
command: /usr/bin/python -c 'import sys, setuptools, tokenize; sys.argv[0] = '"'"'/tmp/pip-install-9DXqvi/networkx/setup.py'"'"'; file='"'"'/tmp/pip-install-9DXqvi/networkx/setup.py'"'"';f=getattr(tokenize, '"'"'open'"'"', open)(file);code=f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, file, '"'"'exec'"'"'))' egg_info --egg-base /tmp/pip-pip-egg-info-kUvbVY
cwd: /tmp/pip-install-9DXqvi/networkx/
Complete output (6 lines):
NetworkX 2.3+ requires Python 3.5 or later (2.7 detected).

xiashiwendao · 2020-05-05T06:41:38Z

Later I change to python3.5 resolve it. I think we should tell user in insightface.ai and modify the introduce in the READ.ME in github, for thru it just get MXNET only support python 2.7. @nttstar

gameliee · 2021-01-26T04:55:50Z

@nttstar can you share retinaface_mnet025_v2 net train code ，and i want modify the net for convert to caffe

I second this, could you kindly provide the model code @nttstar. Thank you!

nttstar · 2021-05-15T08:46:44Z

pip insightface-0.2.0 is ready now, please check the top post.

nadaboulares · 2021-07-06T13:31:41Z

plz how can i test recognition ??

huynhtruc0309 · 2022-04-04T13:34:39Z

How to choose det_size in detection model retinaface? Please help @nttstar

nttstar pinned this issue Sep 5, 2019

nttstar changed the title ~~Hey, pip install insightface and http://insightface.ai~~ pip install insightface and http://insightface.ai Sep 5, 2019

nttstar added the Example label Sep 5, 2019

nttstar changed the title ~~pip install insightface and http://insightface.ai~~ pip install insightface and https://insightface.ai Jul 6, 2021

nttstar closed this as completed Jun 3, 2023

marco-eckstein mentioned this issue Jul 6, 2023

Provide basic end-user documentation #2362

Open

yingfeng unpinned this issue Sep 15, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

pip install insightface and https://insightface.ai #891

pip install insightface and https://insightface.ai #891

nttstar commented Sep 5, 2019 •

edited

Loading

Talgin commented Sep 5, 2019

nttstar commented Sep 5, 2019

Talgin commented Sep 5, 2019

rebotnix commented Sep 6, 2019

RomanSteinberg commented Sep 9, 2019 •

edited

Loading

RomanSteinberg commented Sep 9, 2019

nttstar commented Sep 9, 2019

RomanSteinberg commented Sep 9, 2019

dexception commented Sep 17, 2019

khanhnt commented Oct 10, 2019 •

edited

Loading

ucasiggcas commented Nov 27, 2019 •

edited

Loading

Programmerwyl commented Dec 19, 2019 •

edited

Loading

benouinirachid commented Jan 27, 2020

khanhnt commented Jan 27, 2020

xiashiwendao commented May 4, 2020 •

edited

Loading

xiashiwendao commented May 5, 2020

gameliee commented Jan 26, 2021

nttstar commented May 15, 2021

nadaboulares commented Jul 6, 2021

huynhtruc0309 commented Apr 4, 2022

pip install insightface and https://insightface.ai #891

pip install insightface and https://insightface.ai #891

Comments

nttstar commented Sep 5, 2019 • edited Loading

2021.05.15 Update:

Talgin commented Sep 5, 2019

nttstar commented Sep 5, 2019

Talgin commented Sep 5, 2019

rebotnix commented Sep 6, 2019

RomanSteinberg commented Sep 9, 2019 • edited Loading

RomanSteinberg commented Sep 9, 2019

nttstar commented Sep 9, 2019

RomanSteinberg commented Sep 9, 2019

dexception commented Sep 17, 2019

khanhnt commented Oct 10, 2019 • edited Loading

ucasiggcas commented Nov 27, 2019 • edited Loading

Programmerwyl commented Dec 19, 2019 • edited Loading

benouinirachid commented Jan 27, 2020

khanhnt commented Jan 27, 2020

xiashiwendao commented May 4, 2020 • edited Loading

xiashiwendao commented May 5, 2020

gameliee commented Jan 26, 2021

nttstar commented May 15, 2021

nadaboulares commented Jul 6, 2021

huynhtruc0309 commented Apr 4, 2022

nttstar commented Sep 5, 2019 •

edited

Loading

RomanSteinberg commented Sep 9, 2019 •

edited

Loading

khanhnt commented Oct 10, 2019 •

edited

Loading

ucasiggcas commented Nov 27, 2019 •

edited

Loading

Programmerwyl commented Dec 19, 2019 •

edited

Loading

xiashiwendao commented May 4, 2020 •

edited

Loading