Comments (16)
@Jay-Liu 没没没,我也在学习中,也都是在其他大佬的肩膀上玩,开源就是和大家互相交流学习哈~
from ultra-light-fast-generic-face-detector-1mb.
用640x480输入推理
from ultra-light-fast-generic-face-detector-1mb.
我测的是固定中小输入下的精度,现实场景一般都是固定大小输入,比如320或者640
from ultra-light-fast-generic-face-detector-1mb.
@Linzaer
wider face测试的图片和./imgs目录下的图片不是固定输入大小,detect_img.py run_video_face_detect.py里并没有resize操作而是直接读取图片再predict。
所以我不太明白的是,用640*480的输入推理是怎么在wider face和./imgs目录下的图片进行推理的?
from ultra-light-fast-generic-face-detector-1mb.
@yushanhu12345 有resize的哦,在PredictionTransform里,封装到预处理里了
from ultra-light-fast-generic-face-detector-1mb.
@Linzaer
哦,明白了!
难怪很多大图识别效果不好,准备尝试一下把大图crop出几个vga的tile分别predict,效果应该会好不少
from ultra-light-fast-generic-face-detector-1mb.
@yushanhu12345 可以尝试修改图片输入大小,比如detect_imgs.py里面修改开头第11行的input_img_size,可以适当加大,默认提供(128/160/320/480/640/1280)这几个输入可选。同时如果你想识别大图小人脸场景,建议使用 Mb_Tiny_RFB_FD_train_input_640.pth这个模型,是用640x480训练的,更加适合大分辨率输入,误报低。
from ultra-light-fast-generic-face-detector-1mb.
@Linzaer
自己训练的一些模型和你这个模型,好像都有个问题,就是用大分辨率训练的模型,误检少,对小脸的recall低,
用小分辨率训练的模型,误检多,对小脸的recall高。
刚用Mb_Tiny_RFB_FD_train_input_320.pth和Mb_Tiny_RFB_FD_train_input_640.pth测试过,其他条件都一样(输入640×480的图片,从原图上crop, threshold=0.3)
from ultra-light-fast-generic-face-detector-1mb.
@yushanhu12345 一般来说是都这样的,应该是不同大小输入训练的感受野对应原图大小区域的问题。
from ultra-light-fast-generic-face-detector-1mb.
来围观一下大神@-@
from ultra-light-fast-generic-face-detector-1mb.
可以说说训练这个检测器的心得嘛。效果确实不错,速度也快。
from ultra-light-fast-generic-face-detector-1mb.
@leiyu1980 调参压榨结构,炼丹
from ultra-light-fast-generic-face-detector-1mb.
问一下anchor是如何设置的?
from ultra-light-fast-generic-face-detector-1mb.
@nttstar 大佬好~anchor设计参考了于老师的libfacedetection,四分支3/2/2/3个数配比正方形,加强小人脸和大人脸能力。320x240输入得到的四个feature尺寸(40,20,10,5),然后8*n关系设置大小,每层2倍增加。我也用check_gt_box.py得到训练集的尺寸分布调整过数值,效果差不多。
from ultra-light-fast-generic-face-detector-1mb.
有个疑惑,如上代码是crop出一个h=w的正方形图像,数据扩展也没有padding的过程,那这样在直接resize后人脸会有形变呀。
from ultra-light-fast-generic-face-detector-1mb.
@tfygg 你好,由于人脸主要贴近正方形,所以选的纵横比是1,但是能等于1 的时候几率是非常低的,几乎不会进入下面的randomcrop,所以绝大部分数据几乎是原图,而且如果等于1,resize一定的形变也是可以接受的,因为即使有Expand,当randomcrop出小区域的不同比例的时候,resize之后也是会形变的。由于模型比较小,所以没采用Expand去扩展原图来加强小人脸检出能力,测试调参发现去掉Expand效果会好点,可能是原本数据集的小人脸就已经足够去拟合了。不过这是我的测试结果和想法,你也可以去调参试试~期待更好的结果。
from ultra-light-fast-generic-face-detector-1mb.
Related Issues (20)
- How to increase amount of using CPUs?
- 请问可以批量推理吗
- About the BBox Detection for "masked-Face"
- ModuleNotFoundError: No module named 'vision'
- error building for ncnn
- 请问README中的测试精度是指什么?
- Error while training HOT 1
- Bounding box overlap issue.
- Min_boxes (anchors) calculation
- 请问如何训练灰度图? HOT 1
- ModuleNotFoundError: No module named 'tf' in convert_tensorflow.py HOT 1
- converting to tfjs model
- onnx转换出来报错
- Transfer learning and lable output
- 对全景图进行人脸识别
- 有一段代码不是很理解,有哪位大佬帮我解下惑 HOT 1
- 代码参数理解 HOT 1
- Improve accuracy of the ultraface-rfb-640.onnx model
- 如何优化GPU训练速度
- 大哥你也太强了, 啥都会啊, 这代码没注释根本看不懂
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from ultra-light-fast-generic-face-detector-1mb.