A parent folder named “mhp_parsing_results_template.zip” (Click to download a template file) contains 2 folders and 1 .txt file:
1. results.txt
A plain text file. Each line begins with an image id (as in ./list/ of the LV-MHP-V2 dataset), followed by multiple pairs of person id and confidence score. Person ids range from 1 to the total number of persons, and confidence scores are used to rank the person instances. The separator is a space.
2. global_seg
A folder of png images. For an image id in results.txt, there should be a file named id.png under this folder. The content of id.png is the global person parsing results (instance-agnostic) for the image with exactly the same size. Then id.png file contains integer values 0,1,2,..58, following the same category definition in the README file of LV-MHP-V2.
3. global_tag
A folder of png images. For an image id in results.txt, there should be a file named id.png under this folder. The content of id.png is the person instance map for the image with exactly the same size. Then id.png file contains integer values 0,1,2,..Inf, where 0 indicates background, 1 indicates the person with person id 1 in results.txt, 2 indicates person id 2, and so on.
Make sure these and then package the folder with .zip format. Submit your mhp_parsing_results.zip to mhp.lv.nus@gmail.com with your team information and method abstraction, and wait to see your rank. Note that each team can submit their results for at most three times.
A mat file named "mhp_pose_results_template.mat" (Click to download a template file) with the structure:
{pred:
annorect
annopoints
point
x
y
id
score}
(x,y) denotes the joint coordinate, id denotes the joint index, score denotes the confidence score.
For your convenience, you can prepare your results according to the following steps after downloading the template result file:
1. Load the template result:
load('mhp_pose_results_template.mat', 'pred');
2. For p-th joint of r-th person in i-th image, set the estimation results:pred(i).annorect(r).annopoints.point(p).x = x_pred; pred(i).annorect(r).annopoints.point(p).y = y_pred; pred(i).annorect(r).annopoints.point(p).id = id; pred(i).annorect(r).annopoints.point(p).score = score;3. Save results into mhp_pose_results.mat:
save('mhp_pose_results.mat','pred');
Submit your mhp_pose_results.mat to mhp.lv.nus@gmail.com with your team information and method abstraction, and wait to see your rank. Note that each team can submit their results for at most three times.
Due Date(mm/dd/yyyy): 06/10/2018 23:59 UTC/GMT+0
