Mask R-CNN Versions Save

This release adds:

• The Balloon Color Splash sample, along with dataset and trained weights.
• Convert the last prediction layer from Python to TensorFlow operations.
• Automatic download of COCO weights and dataset.
• Fixes for running on Windows.

Thanks to everyone who made this possible with fixes and pull requests.

Note: COCO weights are not updated in this release. Continue to use the .h5 file from release 2.0.

This release includes updates to improve training and accuracy, and a new MS COCO trained model.

• Remove unnecessary dropout layer
• Reduce anchor stride from 2 to 1
• Increase ROI training mini batch to 200 per image
• Improve computing proposal positive:negative ratio
• Updated COCO training schedule
• Add --logs param to coco.py to set logging directory
• Bug Fix: exclude BN weights from L2 regularization
• Use mean (rather than sum) of L2 regularization for a smoother loss in TensorBoard
• Better compatibility with Python 2.7

The new MS COCO trained weights improve the accuracy compared to the previous weights. These are the evaluation results on the minival dataset:

Evaluate annotation type *bbox*
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.347
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.544
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.377
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.163
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.390
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.486
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.295
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.424
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.433
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.214
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.481
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.601

Evaluate annotation type *segm*
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.296
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.510
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.306
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.128
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.330
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.430
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.258
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.369
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.376
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.173
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.417
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.538

Big thanks to everyone who contributed to this repo. Names are in the commits history.