Skip to content

Latest commit

 

History

History
67 lines (61 loc) · 3.55 KB

RESULTS.md

File metadata and controls

67 lines (61 loc) · 3.55 KB
Raw

Models

  • All models, unless stated are trained with square inputs.
  • All models are trained on coco2017 train split and evaluated on the coco2017 val split.
  • The RetinaNet paper trains the model for ~ 12.7 epochs on the coco2017, this is referred to as 1x schedule, the models listed below are trained for 1x, 3x or 30x schedules.

ResNet50 640x640

 - Backbone            : ResNet50 v1 (ImageNet pretrained weights)
 - Schedule            : 3x
 - Time required       : 50mins
 - System              : v3-32 TPU pod
 - config              : mscoco-retinanet-resnet50-640x640-3x-256
 - weights             : coming soon 

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.377
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.570
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.401
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.177
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.426
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.551
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.315
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.492
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.520
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.278
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.588
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.710

 - Backbone            : ResNet50 v1 (random weight initialization)
 - Schedule            : 30x
 - Time required       : 9h:30min
 - System              : v3-32 TPU pod
 - config              : mscoco-retinanet-resnet50-640x640-30x-256
 - weights             : coming soon 

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.398
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.590
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.425
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.194
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.450
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.570
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.330
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.513
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.540
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.300
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.608
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.731

Training schedules for COCO2017

batch_size=64 (v2-8/v3-8)

name training steps ~epochs
1x 22500 12
3x 67500 36
30x 675000 360
50x 1125000 600

batch_size=256 (v3-32 pod sclice)

name training steps ~epochs
1x 5625 12
3x 16875 36
30x 168750 360
50x 281250 600