JOIN
Get Time
forums   
Search | Watch Thread  |  My Post History  |  My Watches  |  User Settings
View: Flat  | Threaded  | Tree
Previous Thread  |  Next Thread
Clarification of Test Data Processing: Image versus Temporal Scene classification | Reply
This is a request for clarification regarding the desired output and handling of the testing data.


Regarding:
(https://community.topcoder.com/longcontest/?module=ViewProblemStatement&rd=16996&pm=14684)
Section: Output Files


"
Your output must be a text file that describes the object classifications your algorithm makes for all of the images in a test set. You must make a prediction for each bounding box specified in the metadata files of the test set. The file should contain lines formatted like:
<bounding_box_id>,<category>
"


In the case of the test data, say folder: 0007261, we have 8 individual images that represent a temporal collection.

$ ls 0007261/*_rgb.j* | sort
0007261/0007261_0_rgb.jpg
0007261/0007261_0_rgb.json
0007261/0007261_1_rgb.jpg
0007261/0007261_1_rgb.json
0007261/0007261_2_rgb.jpg
0007261/0007261_2_rgb.json
0007261/0007261_3_rgb.jpg
0007261/0007261_3_rgb.json
0007261/0007261_4_rgb.jpg
0007261/0007261_4_rgb.json
0007261/0007261_6_rgb.jpg
0007261/0007261_6_rgb.json
0007261/0007261_7_rgb.jpg
0007261/0007261_7_rgb.json


These should all be collapsed to a single output line for the BBOX ID = 10311 ?

$ grep ID 0007261/*_rgb.json
0007261/0007261_0_rgb.json: "ID": 10311
0007261/0007261_1_rgb.json: "ID": 10311
0007261/0007261_2_rgb.json: "ID": 10311
0007261/0007261_3_rgb.json: "ID": 10311
0007261/0007261_4_rgb.json: "ID": 10311
0007261/0007261_6_rgb.json: "ID": 10311
0007261/0007261_7_rgb.json: "ID": 10311

Please confirm that the intended output is to be a classification of the temporal sequence associated with a bounding box, not classification per image.
Re: Clarification of Test Data Processing: Image versus Temporal Scene classification (response to post by grant_scott) | Reply
Yes, the output should contain one line per each bounding box present in the test data, regardless of the number of images (temporal views of the same scene) the same bounding box appear.
So in the example you gave box ID 10311 should appear only once in your output.
RSS