OV2.0 provide layout as mo parameter for public models (openvinotoolkit#2945)

* provide layout as mo parameter for public models

* update docs for public tf models

* correct layout for aclnet

* fixes

* Update models/public/deeplabv3/model.yml

* provide layout for new models

* turn off layouts for failed models

Author: eaidova

models/public/Sphereface/model.yml

@@ -27,6 +27,7 @@ files:
 model_optimizer_args:
   - --input_shape=[1,3,112,96]
   - --input=data
+  - --layout=data(NCHW)
   - --mean_values=data[127.5,127.5,127.5]
   - --scale_values=data[128.0]
   - --output=fc5

models/public/aclnet-int8/model.yml

@@ -33,6 +33,7 @@ files:
 model_optimizer_args:
   - --input_shape=[1,1,1,16000]
   - --input=result.1
+  - --layout=result.1(NCWL)
   - --output=486
   - --input_model=$dl_dir/aclnet_des_53_int8.onnx
 framework: onnx

models/public/aclnet/model.yml

@@ -31,6 +31,7 @@ files:
 model_optimizer_args:
   - --input_shape=[1,1,1,16000]
   - --input=input
+  - --layout=input(NCWL)
   - --output=output
   - --input_model=$dl_dir/aclnet_des_53.onnx
 framework: onnx

models/public/alexnet/model.yml

@@ -43,6 +43,7 @@ model_optimizer_args:
   - --input_shape=[1,3,227,227]
   - --input=data
   - --mean_values=data[104.0,117.0,123.0]
+  - --layout=data(NCHW)
   - --output=prob
   - --input_model=$dl_dir/alexnet.caffemodel
   - --input_proto=$dl_dir/alexnet.prototxt

models/public/anti-spoof-mn3/model.yml

@@ -32,6 +32,7 @@ model_optimizer_args:
   - --input=actual_input_1
   - --mean_values=actual_input_1[151.2405,119.5950,107.8395]
   - --scale_values=actual_input_1[63.0105,56.4570,55.0035]
+  - --layout=actual_input_1(NCHW)
   - --reverse_input_channels
   - --input_model=$dl_dir/anti-spoof-mn3.onnx
 framework: onnx

models/public/bert-base-ner/model.yml

@@ -89,6 +89,7 @@ conversion_to_onnx_args:
 model_optimizer_args:
   - --input_shape=[1,128],[1,128],[1,128]
   - --input=input_ids,attention_mask,token_type_ids
+  - --layout=input_ids(NC),attention_mask(NC),token_type_ids(NC)
   - --input_model=$conv_dir/bert-base-ner.onnx
   - --output=output
 framework: pytorch

models/public/brain-tumor-segmentation-0001/model.yml

@@ -41,6 +41,7 @@ files:
 model_optimizer_args:
   - --input_shape=[1,4,128,128,128]
   - --input=data_crop
+  - --layout=data_crop(NCDHW)
   - --output=blockgrad7
   - --input_model=$dl_dir/brain-tumor-segmentation-0001-0000.params
 framework: mxnet

models/public/brain-tumor-segmentation-0002/model.yml

@@ -27,6 +27,7 @@ files:
     source: https://storage.openvinotoolkit.org/repositories/open_model_zoo/public/2022.1/brain-tumor-segmentation-0002/brain-tumor-segmentation-0002.onnx
 model_optimizer_args:
   - --input_shape=[1,4,128,128,128]
+  - --layout=NCDHW
   - --input_model=$dl_dir/brain-tumor-segmentation-0002.onnx
 framework: onnx
 license: https://raw.githubusercontent.com/lachinov/brats2019/master/LICENSE

models/public/caffenet/model.yml

@@ -28,6 +28,7 @@ files:
 model_optimizer_args:
   - --input_shape=[1,3,227,227]
   - --input=data
+  - --layout=data(NCHW)
   - --mean_values=data[104.0,117.0,123.0]
   - --output=prob
   - --input_model=$dl_dir/bvlc_reference_caffenet.caffemodel

models/public/cocosnet/model.yml

@@ -125,6 +125,7 @@ model_optimizer_args:
   - --output=exemplar_based_output
   - --mean_values=ref_image[127.5,127.5,127.5]
   - --scale_values=ref_image[127.5,127.5,127.5]
+  - --layout=input_seg_map(NCHW),ref_image(NCHW),ref_seg_map(NCHW)
   - --reverse_input_channels
 framework: pytorch
 license: https://raw.githubusercontent.com/microsoft/CoCosNet/33f98d092407094a15a08b0555d6f5359490cd3e/LICENSE

models/public/colorization-siggraph/model.yml

@@ -56,6 +56,7 @@ conversion_to_onnx_args:
 model_optimizer_args:
   - --input_shape=[1,1,256,256],[1,2,256,256],[1,1,256,256]
   - --input=data_l,user_ab,user_map
+  - --layout=data_l(NCHW),user_ab(NCHW),user_map(NCHW)
   - --output=color_ab
   - --input_model=$conv_dir/colorization-siggraph.onnx
 framework: pytorch

models/public/colorization-v2/model.yml

@@ -63,6 +63,7 @@ conversion_to_onnx_args:
 model_optimizer_args:
   - --input_shape=[1,1,256,256]
   - --input=data_l
+  - --layout=data_l(NCHW)
   - --output=color_ab
   - --input_model=$conv_dir/colorization-v2-eccv16.onnx
 framework: pytorch

models/public/common-sign-language-0001/model.yml

@@ -32,6 +32,7 @@ files:
 model_optimizer_args:
   - --input=input
   - --input_shape=[1,3,8,224,224]
+  - --layout=input(NCDHW)
   - --output=output
   - --mean_values=[123.675, 116.28, 103.53]
   - --scale_values=[58.395, 57.12, 57.375]

models/public/ctdet_coco_dlav0_512/model.yml

@@ -67,6 +67,7 @@ conversion_to_onnx_args:
 model_optimizer_args:
   - --input_shape=[1,3,512,512]
   - --input=input.1
+  - --layout=input.1(NCHW)
   - --mean_values=input.1[104.04, 113.985, 119.85]
   - --scale_values=input.1[73.695, 69.87, 70.89]
   - --output=center_heatmap,width_height,regression

models/public/ctpn/README.md

@@ -35,12 +35,12 @@ Mean values: [102.9801, 115.9465, 122.7717].
 
 ### Converted Model
 
-Image, name: `Placeholder`, shape: `1, 3, 600, 600`, format: `B, C, H, W`, where:
+Image, name: `Placeholder`, shape: `1, 600, 600, 3`, format: `B, H, W, C`, where:
 
 - `B` - batch size
-- `C` - number of channels
 - `H` - image height
 - `W` - image width
+- `C` - number of channels
 
 Expected color order: `BGR`.
 
@@ -64,19 +64,19 @@ Expected color order: `BGR`.
 
 ### Converted Model
 
-1. Detection boxes, name: `rpn_bbox_pred/Reshape_1/Transpose`, shape: `1, 40, 18, 18` contains predicted regions, format: `B, A, H, W`, where:
+1. Detection boxes, name: `rpn_bbox_pred/Reshape_1`, contains predicted regions, in format `B, H, W, A`, where:
 
     - `B` - batch size
-    - `A` - vector of 4\*N coordinates, where N is the number of detected anchors.
     - `H` - image height
     - `W` - image width
+    - `A` - vector of 4\*N coordinates, where N is the number of detected anchors.
 
-2. Probability, name: `Reshape_2/Transpose`, shape: `1, 20, 18, 18`, contains probabilities for predicted regions in a[0,1] range in format `B, A, H, W`, where:
+2. Probability, name: `Reshape_2`, contains probabilities for predicted regions in a [0,1] range in format `B, H, W, A`, where:
 
     - `B` - batch size
-    - `A` - vector of 2\*N class probabilities (0 class for background, 1 class for text), where N is the number of detected anchors.
     - `H` - image height
     - `W` - image width
+    - `A` - vector of 4\*N coordinates, where N is the number of detected anchors.
 
 ## Download a Model and Convert it into Inference Engine Format
 

models/public/ctpn/model.yml

@@ -25,6 +25,7 @@ files:
 model_optimizer_args:
   - --input_shape=[1,600,600,3]
   - --input=Placeholder
+  - --layout=Placeholder(NHWC)
   - --mean_values=Placeholder[102.9801,115.9465,122.7717]
   - --output=Reshape_2,rpn_bbox_pred/Reshape_1
   - --input_model=$dl_dir/ctpn.pb

models/public/deblurgan-v2/model.yml

@@ -75,6 +75,7 @@ model_optimizer_args:
   - --scale_values=blur_image[127.5]
   - --input_shape=[1,3,736,1312]
   - --input=blur_image
+  - --layout=blur_image(NCHW)
   - --output=deblur_image
   - --input_model=$conv_dir/deblurgan-v2.onnx
 framework: pytorch

models/public/deeplabv3/README.md

@@ -34,12 +34,12 @@ Expected color order: `RGB`.
 
 ### Converted Model
 
-Image, name: `mul_1/placeholder_port_1`, shape: `1, 3, 513, 513`, format: `B, C, H, W`, where:
+Image, name: `mul_1/placeholder_port_1`, shape: `1, 513, 513, 3`, format: `B, H, W, C`, where:
 
 - `B` - batch size
-- `C` - number of channels
 - `H` - image height
 - `W` - image width
+- `C` - number of channels
 
 Expected color order: `BGR`.
 

models/public/deeplabv3/model.yml

@@ -30,6 +30,7 @@ model_optimizer_args:
   - --reverse_input_channels
   - --input_shape=[1,513,513,3]
   - --input=1:mul_1
+  # - --layout=1:mul_1(NHWC)
   - --output=ArgMax
   - --input_model=$dl_dir/deeplabv3_mnv2_pascal_train_aug/frozen_inference_graph.pb
 framework: tf

models/public/densenet-121-tf/README.md

@@ -37,12 +37,12 @@ Mean values - [123.68, 116.78, 103.94], scale values - [58.395,57.12,57.375].
 
 ### Converted Model
 
-Image, name: `input_1`, shape: `1, 3, 224, 224`, format: `B, C, H, W`, where:
+Image, name: `input_1`, shape: `1, 224, 224, 3`, format: `B, H, W, C`, where:
 
 - `B` - batch size
-- `C` - number of channels
 - `H` - image height
 - `W` - image width
+- `C` - number of channels
 
 Expected color order: `BGR`.
 

models/public/densenet-121-tf/model.yml

@@ -28,6 +28,7 @@ model_optimizer_args:
   - --reverse_input_channels
   - --input_shape=[1,224,224,3]
   - --input=input_1
+  - --layout=input_1(NHWC)
   - --mean_values=input_1[123.68,116.78,103.94]
   - --scale_values=input_1[58.395,57.12,57.375]
   - --output=StatefulPartitionedCall/densenet121/predictions/Softmax

models/public/densenet-121/model.yml

@@ -31,6 +31,7 @@ files:
 model_optimizer_args:
   - --input_shape=[1,3,224,224]
   - --input=data
+  - --layout=data(NCHW)
   - --mean_values=data[103.94,116.78,123.68]
   - --scale_values=data[58.8235294117647]
   - --output=fc6

models/public/detr-resnet50/model.yml

@@ -82,6 +82,7 @@ model_optimizer_args:
   - --reverse_input_channels
   - --input_shape=[1,3,800,1137]
   - --input=input
+  - --layout=input(NCHW)
   - --mean_values=input[123.675,116.28,103.53]
   - --scale_values=input[58.395,57.12,57.375]
   - --output=boxes,scores

models/public/dla-34/model.yml

@@ -46,6 +46,7 @@ model_optimizer_args:
   - --reverse_input_channels
   - --input_shape=[1,3,224,224]
   - --input=data
+  - --layout=data(NCHW)
   - --mean_values=data[123.675,116.28,103.53]
   - --scale_values=data[58.395,57.12,57.375]
   - --output=prob

models/public/drn-d-38/model.yml

@@ -61,6 +61,7 @@ model_optimizer_args:
   - --reverse_input_channels
   - --input_shape=[1,3,1024,2048]
   - --input=input
+  - --layout=input(NCHW)
   - --output=output
   - --mean_values=[73.975742869,83.660769353,73.175805779]
   - --scale_values=[46.653282963,47.574230671,47.041147921]

models/public/efficientdet-d0-tf/README.md

@@ -37,12 +37,12 @@ Channel order is `RGB`.
 
 ### Converted Model
 
-Image, name - `image_arrays/placeholder_port_0`,  shape - `1, 3, 512, 512`, format is `B, C, H, W`, where:
+Image, name - `image_arrays/placeholder_port_0`,  shape - `1, 512, 512, 3`, format is `B, H, W, C`, where:
 
 - `B` - batch size
-- `C` - channel
 - `H` - height
 - `W` - width
+- `C` - channel
 
 Channel order is `BGR`.
 

models/public/efficientdet-d0-tf/model.yml

@@ -117,6 +117,7 @@ postprocessing:
 model_optimizer_args:
   - --input_shape=[1,512,512,3]
   - --input=image_arrays
+  # - --layout=image_arrays(NHWC)
   - --reverse_input_channels
   - --input_model=$conv_dir/efficientdet-d0_saved_model/efficientdet-d0_frozen.pb
   - --transformations_config=$mo_ext_dir/front/tf/automl_efficientdet.json

models/public/efficientdet-d1-tf/README.md

@@ -37,12 +37,12 @@ Channel order is `RGB`.
 
 ### Converted Model
 
-Image, name: `image_arrays/placeholder_port_0`,  shape: `1, 3, 640, 640`, format is `B, C, H, W`, where:
+Image, name: `image_arrays/placeholder_port_0`,  shape: `1, 640, 640, 3`, format is `B, H, W, C`, where:
 
 - `B` - batch size
-- `C` - channel
 - `H` - height
 - `W` - width
+- `C` - channel
 
 Channel order is `BGR`.
 

models/public/efficientdet-d1-tf/model.yml

@@ -117,6 +117,7 @@ postprocessing:
 model_optimizer_args:
   - --input_shape=[1,640,640,3]
   - --input=image_arrays
+  # - --layout=image_arrays(NHWC)
   - --reverse_input_channels
   - --input_model=$conv_dir/efficientdet-d1_saved_model/efficientdet-d1_frozen.pb
   - --transformations_config=$mo_ext_dir/front/tf/automl_efficientdet.json

models/public/efficientnet-b0-pytorch/model.yml

@@ -67,6 +67,7 @@ model_optimizer_args:
   - --reverse_input_channels
   - --input_shape=[1,3,224,224]
   - --input=data
+  - --layout=data(NCHW)
   - --mean_values=data[123.675,116.28,103.53]
   - --scale_values=data[58.395,57.12,57.375]
   - --output=prob

models/public/efficientnet-b0/README.md

@@ -39,12 +39,12 @@ Channel order is `RGB`.
 
 ### Converted Model
 
-Image, name - `sub/placeholder_port_0`,  shape - `1, 3, 224, 224`, format is `B, C, H, W`, where:
+Image, name - `sub/placeholder_port_0`, shape - `1, 224, 224, 3`, format is `B, H, W, C`, where:
 
 - `B` - batch size
-- `C` - channel
 - `H` - height
 - `W` - width
+- `C` - channel
 
 Channel order is `BGR`.
 

models/public/efficientnet-b0/model.yml

@@ -32,6 +32,7 @@ postprocessing:
 model_optimizer_args:
   - --input_shape=[1,224,224,3]
   - --input=0:sub
+  # - --layout=0:sub(NHWC)
   - --output=logits
   - --input_meta_graph=$dl_dir/efficientnet-b0/model.ckpt.meta
   - --reverse_input_channels

models/public/f3net/model.yml

@@ -63,6 +63,7 @@ conversion_to_onnx_args:
 model_optimizer_args:
   - --input_shape=[1,3,352,352]
   - --input=input.1
+  - --layout=input.1(NCHW)
   - --input_model=$conv_dir/f3net.onnx
   - --mean_values=input.1[124.55,118.90,102.94]
   - --scale_values=input.1[56.77,55.97,57.50]

models/public/face-detection-retail-0044/model.yml

@@ -31,6 +31,7 @@ files:
 model_optimizer_args:
   - --input_shape=[1,3,300,300]
   - --input=data
+  - --layout=data(NCHW)
   - --input_model=$dl_dir/face-detection-retail-0044.caffemodel
   - --input_proto=$dl_dir/face-detection-retail-0044.prototxt
 framework: caffe

models/public/face-recognition-resnet100-arcface-onnx/model.yml

@@ -31,6 +31,7 @@ model_optimizer_args:
   - --input_shape=[1,3,112,112]
   - --input_model=$dl_dir/arcfaceresnet100-8.onnx
   - --input=data
+  - --layout=data(NCHW)
   - --output=fc1
 framework: onnx
 license: https://raw.githubusercontent.com/onnx/models/master/LICENSE

models/public/faceboxes-pytorch/model.yml

@@ -43,6 +43,7 @@ conversion_to_onnx_args:
 model_optimizer_args:
   - --input_shape=[1,3,1024,1024]
   - --input=input.1
+  - --layout=input.1(NCHW)
   - --input_model=$conv_dir/faceboxes-pytorch.onnx
   - --mean_values=input.1[104.0,117.0,123.0]
   - --output=boxes,scores

models/public/facenet-20180408-102900/README.md

@@ -37,12 +37,12 @@ FaceNet: A Unified Embedding for Face Recognition and Clustering. For details se
 
 ### Converted model
 
-Image, name - `image_batch/placeholder_port_0`, shape - `1, 3, 160, 160`, format `B, C, H, W`, where:
+Image, name - `image_batch/placeholder_port_0`, shape - `1, 160, 160, 3`, format `B, H, W, C`, where:
 
 - `B` - batch size
-- `C` - number of channels
 - `H` - image height
 - `W` - image width
+- `C` - number of channels
 
 Expected color order: `BGR`.
 

models/public/facenet-20180408-102900/model.yml

@@ -33,6 +33,7 @@ model_optimizer_args:
   - --reverse_input_channels
   - --input_shape=[1,160,160,3],[1]
   - --input=image_batch,phase_train
+  # - --layout=image_batch(NHWC)
   - --mean_values=image_batch[127.5,127.5,127.5]
   - --scale_values=image_batch[128.0]
   - --output=embeddings

models/public/fast-neural-style-mosaic-onnx/model.yml

@@ -27,6 +27,7 @@ files:
     size: 6728029
 model_optimizer_args:
   - --input_shape=[1,3,224,224]
+  - --layout=NCHW
   - --input_model=$dl_dir/fast-neural-style-mosaic-onnx.onnx
   - --reverse_input_channels
 framework: onnx