原文:
This page contains the test images that we used for our Wallflower paper.* We tested several different background maintenance algorithms on seven different image sequences. For each test image in the sequence, the output of the algorithm was an image segmented into background and foreground pixels.
Each of the seven different test sequences represents a different, potentially problematic scenario for background maintenance. In general we used the first 100 images in each sequence for training. We then ran each algorithm on the remainder of the sequence up to a prespecified evaluation image. We compared the algorithm's result on the evaluation image to a hand-segmented version of the evaluation image.
We have created one .zip file for each of the seven test scenarios. Each .zip file contains the training images, test images, and hand-segmented evaluation image for comparison, all in BMP format. Each .zip file also contains a file called "script.txt" that gives a short description of the events in the sequence and specifies which images are used for training, testing, and evaluation.
The table below shows sample images and gives links to the ftp site with the .zip files.
译文:
本页包含我们Wallflower paper的测试图片。我们对七种不同的图像序列测试了几个不同的背景维护算法。对于每一个测试图像序列,该算法的输出是一个被分割的图像前景和背景。
七个不同的测试序列中每一个都代表一个背景维护算法中不同的,有潜在的问题的场景。一般,我们用前100幅图像序列来进行训练,而用剩余的图像序列来进行评估。我们将评估的图像和手动分割图像进行了比较。
我们已经对七个测试场景中的每一个创建了zip文件。每个zip文件包含训练图像,测试图像以及手动分割的评价图像,所有都是BMP格式。每个zip文件还包含一个“script.txt”文件,具体说明哪些图像序列用来训练,测试,评价。
下表显示样本图像和给的FTP站点链接zip文件。
Scenario | First Training Image | Evaluation Image | Hand-Segmented Evaluation Image | Notes | .zip File |
Moved Object | ![]() | ![]() | ![]() | Chair is moved | MovedObject.zip (79 MB) |
Time of Day | ![]() | ![]() | ![]() | Lights gradually brighten simulating moving sun | TimeOfDay.zip (193 MB) |
Light Switch | ![]() | ![]() | ![]() | Person switches on light | LightSwitch.zip (106 MB) |
Waving Trees | ![]() | ![]() | ![]() | Tree moving in background | WavingTrees.zip (14 MB) |
Camouflage | ![]() | ![]() | ![]() | Static foreground occludes dynamic background | Camouflage.zip (12 MB) |
Bootstrapping | ![]() | ![]() | ![]() | No pure background training available | Bootstrap.zip (145 MB) |
Foreground Aperture | ![]() | ![]() | ![]() | Trained on sleeping person who wakes up | ForegroundAperture.zip (100 MB) |
paper Citation:Kentaro Toyama, John Krumm, Barry Brumitt, Brian Meyers, "Wallflower: Principles and Practice of Background Maintenance", Seventh International Conference on Computer Vision, September 1999, Kerkyra, Greece, pp. 255-261, IEEE Computer Society Press.