Skip to content

Latest commit

 

History

History
299 lines (232 loc) · 10.3 KB

validation_results_tensorflow.md

File metadata and controls

299 lines (232 loc) · 10.3 KB
Raw

Validation results for the models inferring using Intel® Optimizations for TensorFlow

Image classification

Test image #1

Data source: ImageNet

Image resolution: 709 x 510

Model Parameters Python API
densenet-121-tf - -
efficientnet-b0 - -
googlenet-v1-tf - -
googlenet-v2-tf - -
googlenet-v3 - -
googlenet-v4-tf - -
inception-resnet-v2-tf Channel order is RGB.
Mean values - [127.5, 127.5, 127.5],
scale value - 127.5.		 9.1747866 Granny Smith
4.0729303 pomegranate
3.7423978 orange
3.7375512 bell pepper
3.6937847 piggy bank, penny bank
mixnet-l - -
mobilenet-v1-1.0-224-tf Channel order is RGB.
Mean values - [127.5, 127.5, 127.5],
scale value - 127.5.		 0.1775393 necklace
0.1625960 saltshaker, salt shaker
0.0680758 pitcher, ewer
0.0600448 syringe
0.0574061 Granny Smith
mobilenet-v2-1.0-224 Channel order is RGB.
Mean values - [127.5, 127.5, 127.5],
scale value - 127.5.		 0.8931151 Granny Smith
0.0335338 piggy bank, penny bank
0.0027360 saltshaker, salt shaker
0.0021255 vase
0.0016607 pitcher, ewer
mobilenet-v2-1.4-224 Channel order is RGB.
Mean values - [127.5, 127.5, 127.5],
scale value - 127.5.		 0.7240402 Granny Smith
0.0312107 vase
0.0237109 fig
0.0122461 piggy bank, penny bank
0.0118888 saltshaker, salt shaker
mobilenet-v3-small-1.0-224-tf - -
mobilenet-v3-large-1.0-224-tf - -
resnet-50-tf Channel order is RGB.
Mean values - [123.68, 116.78, 103.94]		 0.9553044 Granny Smith
0.0052123 lemon
0.0047184 piggy bank, penny bank
0.0045875 orange
0.0044232 necklace

Test image #2

Data source: ImageNet

Image resolution: 500 x 500

Model Parameters Python API
densenet-121-tf - -
efficientnet-b0 - -
googlenet-v1-tf - -
googlenet-v2-tf - -
googlenet-v3 - -
googlenet-v4-tf - -
inception-resnet-v2-tf Channel order is RGB.
Mean values - [127.5, 127.5, 127.5],
scale value - 127.5.		 10.2994785 junco, snowbird
5.9667974 brambling, Fringilla montifringilla
3.8809638 indigo bunting, indigo finch, indigo bird, Passerina cyanea
3.7881403 house finch, linnet, Carpodacus mexicanus
3.4699843 goldfinch, Carduelis carduelis
mixnet-l - -
mobilenet-v1-1.0-224-tf Channel order is RGB.
Mean values - [127.5, 127.5, 127.5],
scale value - 127.5.		 0.9818491 junco, snowbird
0.0097170 house finch, linnet, Carpodacus mexicanus
0.0029993 brambling, Fringilla montifringilla
0.0022394 goldfinch, Carduelis carduelis
0.0022212 chickadee
mobilenet-v2-1.0-224 Channel order is RGB.
Mean values - [127.5, 127.5, 127.5],
scale value - 127.5.		 0.8770270 junco, snowbird
0.0143872 water ouzel, dipper
0.0103318 chickadee
0.0063065 brambling, Fringilla montifringilla
0.0013868 red-backed sandpiper, dunlin, Erolia alpina
mobilenet-v2-1.4-224 Channel order is RGB.
Mean values - [127.5, 127.5, 127.5],
scale value - 127.5.		 0.6637316 junco, snowbird
0.0811651 chickadee
0.0119593 water ouzel, dipper
0.0038528 brambling, Fringilla montifringilla
0.0022498 goldfinch, Carduelis carduelis
mobilenet-v3-small-1.0-224-tf - -
mobilenet-v3-large-1.0-224-tf - -
resnet-50-tf Channel order is RGB.
Mean values - [123.68, 116.78, 103.94]		 0.9983400 junco, snowbird
0.0004680 brambling, Fringilla montifringilla
0.0003848 chickadee
0.0003656 water ouzel, dipper
0.0003383 goldfinch, Carduelis carduelis

Test image #3

Data source: ImageNet

Image resolution: 333 x 500

Model Parameters Python API
densenet-121-tf - -
efficientnet-b0 - -
googlenet-v1-tf - -
googlenet-v2-tf - -
googlenet-v3 - -
googlenet-v4-tf - -
inception-resnet-v2-tf Channel order is RGB.
Mean values - [127.5, 127.5, 127.5],
scale value - 127.5.		 6.6930799 fireboat
6.1025167 breakwater, groin, groyne, mole, bulwark, seawall, jetty
6.0896273 lifeboat
5.7389712 container ship, containership, container vessel
5.4940562 dock, dockage, docking facility
mixnet-l - -
mobilenet-v1-1.0-224-tf Channel order is RGB.
Mean values - [127.5, 127.5, 127.5],
scale value - 127.5.		 0.3759801 liner, ocean liner
0.1252522 lifeboat
0.1200093 breakwater, groin, groyne, mole, bulwark, seawall, jetty
0.0882490 beacon, lighthouse, beacon light, pharos
0.0568063 fireboat
mobilenet-v2-1.0-224 Channel order is RGB.
Mean values - [127.5, 127.5, 127.5],
scale value - 127.5.		 0.1885883 beacon, lighthouse, beacon light, pharos
0.1434043 liner, ocean liner
0.0768170 breakwater, groin, groyne, mole, bulwark, seawall, jetty
0.0497303 drilling platform, offshore rig
0.0225758 container ship, containership, container vessel
mobilenet-v2-1.4-224 Channel order is RGB.
Mean values - [127.5, 127.5, 127.5],
scale value - 127.5.		 0.1300134 container ship, containership, container vessel
0.0765783 lifeboat
0.0406071 dock, dockage, docking facility
0.0393021 drilling platform, offshore rig
0.0381023 liner, ocean liner
mobilenet-v3-small-1.0-224-tf Channel order is RGB. -
mobilenet-v3-large-1.0-224-tf Channel order is RGB. -
resnet-50-tf Channel order is RGB.
Mean values - [123.68, 116.78, 103.94]		 0.2357705 breakwater, groin, groyne, mole, bulwark, seawall, jetty
0.1480758 liner, ocean liner
0.1104694 container ship, containership, container vessel
0.1095414 drilling platform, offshore rig
0.0915567 beacon, lighthouse, beacon light, pharos

Object detection

Test image #1

Data source: ImageNet

Image resolution: 709 x 510

Bounding boxes (upper left and bottom right corners):
(55, 155), (236, 375)
(190, 190), (380, 400)
(374, 209), (588, 422)
(289, 111), (440, 255)
(435, 160), (615, 310)
Model Python API
ctpn -
efficientdet-d0 -
efficientdet-d1 -
faster_rcnn_inception_resnet_v2_atrous_coco -
faster_rcnn_resnet50_coco -
retinanet -
rfcn-resnet101-coco -
ssd_mobilenet_v1_coco -
ssd_mobilenet_v1_fpn_coco -
ssdlite_mobilenet_v2 -

Test image #2

Data source: ImageNet

Image resolution: 500 x 500

Bounding box (upper left and bottom right corners):
(117, 86), (365, 465)
Model Python API
ctpn -
efficientdet-d0 -
efficientdet-d1 -
faster_rcnn_inception_resnet_v2_atrous_coco -
faster_rcnn_resnet50_coco -
retinanet -
rfcn-resnet101-coco -
ssd_mobilenet_v1_coco -
ssd_mobilenet_v1_fpn_coco -
ssdlite_mobilenet_v2 -

Test image #3

Data source: ImageNet

Image resolution: 333 x 500

Bounding box (upper left and bottom right corners):
(82, 262), (269, 376)
Model Python API
ctpn -
efficientdet-d0 -
efficientdet-d1 -
faster_rcnn_inception_resnet_v2_atrous_coco -
faster_rcnn_resnet50_coco -
retinanet -
rfcn-resnet101-coco -
ssd_mobilenet_v1_coco -
ssd_mobilenet_v1_fpn_coco -
ssdlite_mobilenet_v2 -

Test image #4

Data source: MS COCO

Image resolution: 640 x 480

Bounding boxes (upper left and bottom right corners):
TV (110, 41), (397, 304)
MOUSE (508, 337), (559, 374)
KEYBOARD (241, 342), (496, 461)
Model Python API
ctpn -
efficientdet-d0 -
efficientdet-d1 -
faster_rcnn_inception_resnet_v2_atrous_coco -
faster_rcnn_resnet50_coco -
retinanet -
rfcn-resnet101-coco -
ssd_mobilenet_v1_coco -
ssd_mobilenet_v1_fpn_coco -
ssdlite_mobilenet_v2 -

Test image #5

Data source: Pascal VOC

Image resolution: 500 x 375

Bounding box (upper left and bottom right corners):
AEROPLANE (131, 21), (248, 414)
Model Python API
ctpn -
efficientdet-d0 -
efficientdet-d1 -
faster_rcnn_inception_resnet_v2_atrous_coco -
faster_rcnn_resnet50_coco -
retinanet -
rfcn-resnet101-coco -
ssd_mobilenet_v1_coco -
ssd_mobilenet_v1_fpn_coco -
ssdlite_mobilenet_v2 -

Test image #6

Data source: MS COCO

Image resolution: 640 x 427

Bounding boxes (upper left and bottom right corners):
PERSON (86, 84), (394, 188)
HORSE (44, 108), (397, 565)

Model Python API
ctpn -
efficientdet-d0 -
efficientdet-d1 -
faster_rcnn_inception_resnet_v2_atrous_coco -
faster_rcnn_resnet50_coco -
retinanet -
rfcn-resnet101-coco -
ssd_mobilenet_v1_coco -
ssd_mobilenet_v1_fpn_coco -
ssdlite_mobilenet_v2 -

Semantic segmentation

Test image #1

Data source: -

Image resolution: -

Image: -

Segmented images are identical.

Model Python API
deeplabv3 -

Instance segmentation

Test image #1

Data source: MS COCO

Image resolution: 640 x 480

Image:

Segmented images are identical.

Model Python API
mask_rcnn_resnet50_atrous_coco -
mask_rcnn_inception_resnet_v2_atrous_coco -

Color map: