mAP = nan

[vivva]

Hello !

I only used the first four categories for training and testing, taking four videos for each category. Due to the limitation of computer, I can't train the SSN model. So I directly use the reference model to extract the test score of the proposal. When I run

'python ssn_test.py thumos14 RGB None bninception_thumos14_imagenet_rgb_2.scores.npz --use_reference'
I set num class to 20 to get the test score.

But when I run

python eval_detection_results.py thumos14 bninception_thumos14_imagenet_rgb_2.scores.npz
it shows up:

If I set num class to 4 and run '`python eval_detection_results.py thumos14 bninception_thumos14_imagenet_rgb_2.scores.npz
' again, there will be a size mismatch error.

How can I solve it?

[vivva]

[vivva]

it shows up:
Performing nms...
NMS Done.
Performing location regression...
Regression Done.
Calling mean AP calculator from toolkit with 32 workers...
Evaluation done.

+Detection Performance on thumos14+------+------+------+------+------+------+---------+
| IoU thresh | 0.10 | 0.20 | 0.30 | 0.40 | 0.50 | 0.60 | 0.70 | 0.80 | 0.90 | Average |
+------------+------+------+------+------+------+------+------+------+------+---------+
| mean AP | nan | nan | nan | nan | nan | nan | nan | nan | nan | nan |
+------------+------+------+------+------+------+------+------+------+------+---------+