Training hardware list?

[zhousaYolo]

When I use 1080ti, the parameter batchsize = 16, on the coco data set, I can't restore your training results.

[610265158]

Do you mean you trained the model, but the result is not well?
what is your map then?

[zhousaYolo]

Yes ,The Total loss is about 2.8.Can't converge.

[610265158]

For mscoco, the final loss with mbv3 is about 3. 5 , there should be something wrong.
Please open vis config, to check the data

[zhousaYolo]

Thank you! Do you mean that for mscoco dataset, the total loss of final training is about 3.5, so it can be considered as convergence? its [email protected] Can it be accurate to 0.4? I mistakenly think that the total loss should be less than 1 before it is considered to be convergent and can be evaluated.

In addition, if you use shufflernet as backone, can it converge? [email protected] How much is it?

[zhousaYolo]

I use your “ detector.pb ", testing the small mscoco data set (6 pictures), it is found that objects can be detected, but in calculating the map value, it is all 0. What's the reason?

Accumulating evaluation results...

DONE (t=0.00s).

Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.000

Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.000

Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.000

Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000

Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.000

Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.000

Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.000

Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.000

Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.000

Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000

Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.000

Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.000

[610265158]

Better do visulization the result first,
Make the json as cocostyle, and in the right category.