2018/12/04 Fixed bugs in the evaluation code. Please post the result again if you have already posted one.
Bounding boxes = (x1, y1, x2, y2) are assigned to areas where objects to be predicted are present in images taken vehicle front camera, and one category is be assigned to each area. One or more bounding boxes are defined for each image by establishing the upper-left corner of the image as the origin (coordinates 0,0) and specifying four factors: its upper-left coordinates (x1, y1), and the lower-right coordinates (x2, y2) of the bounding box(es). The following six categories are used for objects to be utilized in prediction:
Route: Tokyo Route 1 (Shibuya-Haneda)）,Tokyo Route 2 (Kagurazaka), Saitama Route (Bijogi)
Time of day: 3 time categories of "morning," "day" and "night"; only "day" is used for Tokyo Route 2
Photograph Intervals: 1 seconds(varies in some cases)
Resolution: 1936 x 1216
No. of images: （Train）21,258（Test）6,355
No. of categories: 10
Category names: Car, Pedestrian, Truck, Signal, Signs, Bicycle, Bus, SVehicle, Motorbike, Train
About the idea award
The idea award is opened for those who submitted a report. In response to the theme of "Edge AI", please summarize the explanation of the model and the ideas that are oriented towards the implementation, etc., and then submit it from the "Send Report" button.
・Even though you are willing to challenge to only the idea award other than the accuracy award,, it is necessary to create models and post prediction results.
・You can choose any type of the form of the report, but please describe the segmentation division, title, and your user name on SIGNATE on the 1st page in it.
・Please write the following contents in the report.
- How to use and process data
- Points oriented towards implementation in building a model, such as memory capacity and processing speed.
- Consideration on the prediction result (for example, a discussion of the relationship between speed and accuracy, an analysis on experimentally verified results, and so forth.)
・After printing the report to pdf with a file name of "(object_detection) _ (username). pdf", please submit it.
・You can submit the report more than once, but note that the last posted file for each assignment will be subject to evaluation.