some questions in evalCOCO.m

[louielu1027]

@ZheC Hi, when i ran evalCOCO.m ,i got an error

Undefind function or variable 'coco_val'.

So, what does the 'coco_val' mean in evalCOCO.m? How can i solve this problem? Thank you very much!

[xiaojieli0903]

I'm also confused about this question.

[louielu1027]

I don't use this code. Instead, I use the code in COCO toolbox(MatlabAPI/evalDemo.m), the json file is generated by examples/tests/pose_accuracy_coco_test.sh ,you need change the json file path in evalDemo.m.........But,I also get another problem, when I generate json file I set ' --frame_last' to 2644 and 3559 ,I get the same AP accuracy......
why ? @ZheC @huaer9258

[xiaojieli0903]

What AP accuravy did you get? Because when I use your methods I got poor AP values about 0.37. But when I solved the problem you proposed and use ZheC's testing code to generate the result.json. I got the nearly same AP value at around 0.573 with the owner's result.

[louielu1027]

I got 0.387 at sacle_number=1 , 0.483 at sacle_number=4 And how do you solve the problem? Thank you very much! @huaer9258

[louielu1027]

@huaer9258 Could you please tell me how to solve the problem? Thank you very much!

[xiaojieli0903]

I just generated a txt including the image name and image id on each line. And read this txt in evalCOCO.m when it needs these information. You can have a try. Hope this method can help you.

[bazhijing]

have you solve the problem？ i got 0.48 ap too。just use coco toolbox to evaluete the json output

[xxxyyyzzzz]

@louielu1027 I couldn't locate examples/tests/pose_accuracy_coco_test.sh file which generates json file. Also will caffemodel be the input for that shell script?

[YanYan0716]

@louielu1027 have you solved the problem about the question. I just did what the 'readme' said to train, and owned my model. but i got 0.43 ap , so i don't know why