liruilong940607 / ochumanapi

API for the dataset proposed in "Pose2Seg: Detection Free Human Instance Segmentation" @ CVPR2019.

OCHuman(Occluded Human) Dataset Api

Dataset proposed in "Pose2Seg: Detection Free Human Instance Segmentation" [ProjectPage] [arXiv] @ CVPR2019.

• News! 2019.06.14 Bug fixed: Val/Test annotation split is now matched to our paper, please update!
• News! 2019.04.08 Codes for our paper is available now!

Samples of OCHuman Dataset

This dataset focus on heavily occluded human with comprehensive annotations including bounding-box, humans pose and instance mask. This dataset contains 13360 elaborately annotated human instances within 5081 images. With average 0.573 MaxIoU of each person, OCHuman is the most complex and challenging dataset related to human. Through this dataset, we want to emphasize occlusion as a challenging problem for researchers to study.

Statistics

All the instances in this dataset are annotated by bounding-box. While not all of them have the keypoint/mask annotation. If you want to compare your results with ours in the paper, please use the subset that contains both keypoint and mask annotations (4731 images, 8110 persons).

	bbox	keypoint	mask	keypoint&mask	bbox&keypoint&mask
#Images	5081	5081	4731	4731	4731
#Persons	13360	10375	8110	8110	8110
#mMaxIou	0.573	0.670	0.669	0.669	0.669

Note:

• MaxIoU measures the severity of an object being occluded, which means the max IoU with other same category objects in a single image.
• All instances in OCHuman with kpt/mask annotations are suffered by heavy occlusion. (MaxIou > 0.5)

Download Links

• Images (667MB) & Annotations

In the above link, we also provide the coco style annotations (val and test subset) so that you can run evaluation using cocoEval toolbox.

Update at 2019.06.14: Please download annotation files (*json) again to match the val/test split used in our paper.

Install API

git clone https://github.com/liruilong940607/OCHumanApi
cd OCHumanApi
make install

How to use

See Demo.ipynb

ochumanapi's Issues

Question on evaluation protocol used in paper

I would like to check some details in your paper "Pose2Seg: Detection Free Human Instance Segmentation". For the AP scores reported in Tables 2 & 3, can I check if these scores are evaluated on what type of IOU computation (bounding box or segmentation)? To be specific, I am refering to the iouType here https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocotools/cocoeval.py

Can not open the download link. 打不开下载链接!

Why only some people are marked?

Why only some people are marked, others in a picture are not marked?

COCO style train annotations

Hello,
Thank you for providing this great dataset!
Do you also provide COCO-style train annotations? Or could you provide me the Code, that you used to transform the annotations to COCO style?

Some questions about mMaxIou

Can MaxIou be considered as similar as visibility score(how much of an object is visible)?
Can you please share the codes to calculate max-IOU between bounding boxes of all pedestrians in an image? Thanks.

Can you share the label tool?

Unable to read test or val annotations using the code given in demo.ipynb

I use the following code:

import os
import cv2
import json
from ochumanApi.ochuman import OCHuman

data_dir = './images'
annot_file = './ochuman.json'
annot_file = './ochuman_coco_format_val_range_0.00_1.00.json'

ochuman = OCHuman(AnnoFile=annot_file, Filter='kpt&segm')
image_ids = ochuman.getImgIds()
print ('Total images: %d'%len(image_ids))


data = ochuman.loadImgs(imgIds=[image_ids[2]])[0]
img = cv2.imread(os.path.join(data_dir, data['file_name']))
height, width = data['height'], data['width']

But it gives the error

Traceback (most recent call last):
  File "visualize_val.py", line 10, in <module>
    ochuman = OCHuman(AnnoFile=annot_file, Filter='kpt&segm')
  File "/usr/local/lib/python3.6/dist-packages/ochumanApi/ochuman.py", line 53, in __init__
    self.keypoint_names = self.dataset['keypoint_names']
KeyError: 'keypoint_names

License of the data

Hi and thank you for this dataset. Could you please specify whether the dataset can be used:

• for non-academic research?
• to train models for commercial applications?

license of each image

Thank you for your wonderful work!
Could you tell me how to know the license of each image?
Thank you.

Source of images

Hi! Thank you for the dataset, it is very useful. Can I check where are the images sourced from? I can't find that info in your paper.

Error while install the API

Hello. While trying to install the API using the method mentioned, i am getting the following error. can you kindly help

error: Unable to find vcvarsall.bat

will the code for pose2seg be available?