yolo

目标检测综述(1)

限于喜欢 提交于 2019-12-16 06:07:44
《最近在做目标检测,差不多懂了。所以看了下综述》 https://zhuanlan.zhihu.com/p/33981103 目标检测一直是计算机视觉的基础问题,在 2010 年左右就开始停滞不前了。自 2013 年一篇论文的发表,目标检测从原始的传统手工提取特征方法变成了基于卷积神经网络的特征提取,从此一发不可收拾。 本文将跟着历史的潮流,简要地探讨「目标检测」算法的两种思想和这些思想引申出的算法,主要涉及那些主流算法,no bells and whistles. 概述 Overview 在深度学习正式介入之前,传统的「目标检测」方法都是 区域选择 、 提取特征 、 分类回归 三部曲,这样就有两个难以解决的问题;其一是区域选择的策略效果差、时间复杂度高;其二是手工提取的特征鲁棒性较差。 云计算时代来临后,「目标检测」算法大家族主要划分为两大派系,一个是 R-CNN 系两刀流,另一个则是以 YOLO 为代表的一刀流派。下面分别解释一下 两刀流 和 一刀流 。 两刀流 顾名思义,两刀解决问题: 1、生成可能区域(Region Proposal) & CNN 提取特征 2、放入分类器分类并修正位置 这一流派的算法都离不开 Region Proposal ,即是优点也是缺点,主要代表人物就是 R-CNN 系。 一刀流 顾名思义,一刀解决问题,直接对 预测的目标物体进行回归 。

Pressure to Images and Images to pressure conversion

烂漫一生 提交于 2019-12-13 04:19:01
问题 EDIT:I am working on the Pressure map dataset where the pressure sensors data is "in-bed posture pressure data". Dataset : https://physionet.org/content/pmd/1.0.0/ Using the below code I could able to convert the pressure data to images. line = f.readlines()[3] lst1 = line.strip().split() lst = [int(x) for x in lst1] # Convert into a 64*32 array rr = np.asarray(lst).reshape(64, 32) plt.imshow(arr, cmap='hot', interpolation='nearest') Images formed are as below: Now, my major motto is to

AttributeError: 'PhotoImage' object has no attribute '_PhotoImage__photo'

痞子三分冷 提交于 2019-12-12 19:24:08
问题 I am working on Yolo3-4-PY to implement it with tkinter. I've looked up everywhere but not able to resolve the issue. When I run the program the canvas is displayed but when I click on Start Video( btton ) I get the following error: Loading weights from weights/yolov3.weights...Done! /usr/local/lib/python3.5/dist-packages/PIL/ImageTk.py:119: FutureWarning: elementwise comparison failed; returning scalar instead, but in the future will perform elementwise comparison if mode not in ["1", "L",

Questions about loss function in yolov2?

半世苍凉 提交于 2019-12-12 01:25:38
问题 I read the yolov2 implementation.I have some questions about it's loss.Below is the pseudo code of the loss function, i hope i got it right. costs = np.zeros(output.shape) for pred_box in all prediction box: if (max iou pred_box has with all truth box < threshold): costs[pred_box][obj] = (sigmoid(obj)-0)^2 * 1 else: costs[pred_box][obj] = 0 costs[pred_box][x] = (sigmoid(x)-0.5)^2 * 0.01 costs[pred_box][y] = (sigmoid(y)-0.5)^2 * 0.01 costs[pred_box][w] = (w-0)^2 * 0.01 costs[pred_box][h] = (h

How do I make the bounding boxes in yolo v3 tighter (closer to the objects)?

匆匆过客 提交于 2019-12-11 17:20:16
问题 I'm following this Repo on creating Yolo v3 model from scratch in PyTorch. The only problem is that the bounding boxes are not as tight (close to the objects) in most images I tried. I compared them to the tutorial on creating Yolo v3 model but using TensorFlow. The tensorflow model produces excellent bounding boxed that are as tight as possible to the objects. I tried to understand how the calculations are different between the two, but I'm finding myself getting stuck with the differences

setup.py egg_info error code 3221225477

一个人想着一个人 提交于 2019-12-11 15:42:10
问题 I've been trying to install IMGAUG package for an ML project. But the installation gets stuck when it tries to install scikit-image My input: pip install scikit-image output: Collecting imgaug Using cached https://files.pythonhosted.org/... Requirement already satisfied: scipy in c:\users\*<username>*\appdata\local\programs\python\python37\lib\site- packages (from imgaug) (1.1.0) Collecting scikit-image>=0.11.0 (from imgaug) Using cached https://files.pythonhosted.org/packages/... Complete

Some tensor manipulations in Tensorflow

不羁的心 提交于 2019-12-11 07:03:00
问题 I've been stuck for several hours on some tensor manipulations in Tensorflow. I'm trying to implement some loss function within the Keras framework using Tensorflow backend, but was unsuccessful so far (though I believe I'm getting closer). The loss function in question is that of the loss function of YOLO algorithm. The problem with this function, is that it is a $L_2$ norm but weighted by some coefficients depending on the value of $y_true$. Therefore, in order to build my loss function, I

【YOLO-V1】You Only Look Once: Unified, Real-Time Object Detection

天涯浪子 提交于 2019-12-10 09:34:38
You Only Look Once: Unified, Real-Time Object Detection 您只看一次:统一的实时对象检测 前言 近几年来,目标检测算法取得了很大的突破。比较流行的算法可以分为两类,一类是基于Region Proposal的R-CNN系算法(R-CNN,Fast R-CNN, Faster R-CNN),他们是two-stage的,需要先使用启发式方法(selective search)或者CNN网络(RPN)产生Region Proposal,然后再在Region proposal上做分类与回归。而另一类是Yolo,SSD这类one-stage算法,其仅仅使用一个CNN网络直接预测不同目标的类别与位置。第一类方法是准确率高一些,但是速度慢,但是第二类算法是速度快,但是准确率要低一些。这里我们谈的是Yolo-v1版本算法,其性能是差于后来的SSD算法的,但是Yolo后来也继续进行改进,产生了Yolo9000算法。本文主要讲述Yolo-v1算法的原理,特别是算法的训练与预测中详细细节,最后将给出如何使用TensorFlow实现Yolo算法。 滑动窗口与CNN 在介绍Yolo算法之前,首先先介绍一下滑动窗口技术,这对我们理解Yolo算法是有帮助的。采用滑动窗口的目标检测算法思路非常简单,它将检测问题转化为了图像分类问题

Darknet YOLO image size

时光毁灭记忆、已成空白 提交于 2019-12-09 16:26:23
问题 I am trying to train custom object classifier in Darknet YOLO v2 https://pjreddie.com/darknet/yolo/ I gathered a dataset for images most of them are 6000 x 4000 px and some lower resolutions as well. Do I need to resize the images before training to be squared ? I found that the config uses: [net] batch=64 subdivisions=8 height=416 width=416 channels=3 momentum=0.9 decay=0.0005 angle=0 saturation = 1.5 exposure = 1.5 hue=.1 thats why I was wondering how to use it for different sizes of data