print pretty evaluation results

[Johnqczhang]

Hi, I make a little improvement on the logging final evaluation results to make it pretty shown in the terminal, like the style of Detectron. A member variable name is added to class COCODataset and PascalVOCDataset respectively to print the dataset name as well.

The difference for an evaluation on a COCO dataset is shown below,

• Before:

2019-06-14 16:07:48,853 maskrcnn_benchmark.inference: 63 INFO: OrderedDict([('bbox', OrderedDict([('AP', 0.5359167831416695), ('AP50', 0.8276995117961133), ('AP75', 0.5836435926893688), ('APs', 0.3651809329419521), ('APm', 0.6116254313370
206), ('APl', 0.6981901486909216)])), ('keypoints', OrderedDict([('AP', 0.6421389106709793), ('AP50', 0.862267027540821), ('AP75', 0.696589535122375), ('APm', 0.5861582705697495), ('APl', 0.7319674179840127)]))])
(base)

• After:

2019-06-14 16:07:48,853 maskrcnn_benchmark.inference: 63 INFO: Dataset: keypoints_coco_2014_minival
Task: bbox
AP, AP50, AP75, APs, APm, APl
0.5359, 0.8277, 0.5836, 0.3652, 0.6116, 0.6982
Task: keypoints
AP, AP50, AP75, APm, APl
0.6421, 0.8623, 0.6966, 0.5862, 0.7320

[fmassa]

Hi,

Thanks for the PR!

There are two changes in this PR:

• add the name attribute to the Datasets
• make the CocoResult structure nicer.

I'm not sure that we should have the name attribute to the datasets, but I do like the result of the new printing in the CocoResult.

What about the following, can you revert the name attribute, and keep the change to the repr?

[Johnqczhang]

Hi @fmassa, adding the name attribute to the datasets is just for printing the dataset name before the evaluation results like the format in Detectron (as shown in the first line of the new printing). It's ok if you think it's unnecessary to add it.

So, how should I do to revert the name and keep the change to the repr?

[Johnqczhang]

Hi @fmassa, I've removed name attribute and only kept the change to the repr.

[fmassa]

Thanks!