Skip to content

Commit 0330712

Browse files
author
Roman Donchenko
authored
Merge pull request openvinotoolkit#1257 from vladimir-dudnik/vd/2020.4-doc-update
fix docs
2 parents 6161647 + 545d921 commit 0330712

File tree

111 files changed

+701
-246
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

111 files changed

+701
-246
lines changed

‎models/intel/index.md

+1-1
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
# Overview of OpenVINO™ Toolkit Pre-Trained Models
1+
# Overview of OpenVINO™ Toolkit Intel's Pre-Trained Models
22

33
OpenVINO™ toolkit provides a set of pre-trained models
44
that you can use for learning and demo purposes or for developing deep learning

‎models/public/Sphereface/Sphereface.md

+4
Original file line numberDiff line numberDiff line change
@@ -17,6 +17,10 @@
1717

1818
## Accuracy
1919

20+
| Metric | Value |
21+
| ------ | ----- |
22+
| LFW accuracy | 98.8321%|
23+
2024
## Performance
2125

2226
## Input

‎models/public/alexnet/alexnet.md

+5
Original file line numberDiff line numberDiff line change
@@ -21,6 +21,11 @@ The model output for `alexnet` is the usual object classifier output for the 100
2121

2222
## Accuracy
2323

24+
| Metric | Value |
25+
| ------ | ----- |
26+
| Top 1 | 56.598% |
27+
| Top 5 | 79.812% |
28+
2429
See [the original model's documentation](https://github.com/BVLC/caffe/tree/master/models/bvlc_alexnet).
2530

2631
## Performance

‎models/public/brain-tumor-segmentation-0001/brain-tumor-segmentation-0001.md

+18-4
Original file line numberDiff line numberDiff line change
@@ -2,9 +2,9 @@
22

33
## Use Case and High-Level Description
44

5-
This model was created for participation in the [Brain Tumor Segmentation Challenge](https://www.med.upenn.edu/sbia/brats2018.html) (BraTS) 2018.
6-
The model is based on [the corresponding paper](https://arxiv.org/abs/1810.04008), where authors present deep cascaded approach for automatic brain tumor segmentation. The paper describes modifications to 3D UNet architecture and specific augmentation strategy to efficiently handle multimodal MRI input. Besides this, the approach to enhance segmentation quality with context obtained from models of the same topology operating on downscaled data is introduced.
7-
Each input modality has its own encoder which are later fused together to produce single output segmentation.
5+
This model was created for participation in the [Brain Tumor Segmentation Challenge](https://www.med.upenn.edu/sbia/brats2018.html) (BraTS) 2018.
6+
The model is based on [the corresponding paper](https://arxiv.org/abs/1810.04008), where authors present deep cascaded approach for automatic brain tumor segmentation. The paper describes modifications to 3D UNet architecture and specific augmentation strategy to efficiently handle multimodal MRI input. Besides this, the approach to enhance segmentation quality with context obtained from models of the same topology operating on downscaled data is introduced.
7+
Each input modality has its own encoder which are later fused together to produce single output segmentation.
88

99
## Example
1010

@@ -19,14 +19,28 @@ Each input modality has its own encoder which are later fused together to produc
1919

2020
## Accuracy
2121

22+
The following accuracy metrics are measured on a `brain tumor` training subset of the [Medical Decathlon](http://medicaldecathlon.com/) dataset.
23+
24+
**Mean**:
25+
- Dice index for "overall": 92.4003%
26+
- Dice index for "necrotic core / non-enhancing tumor": 71.467%
27+
- Dice index for "edema": 82.0533%
28+
- Dice index for "enhancing tumor": 72.7001%
29+
30+
**Median**:
31+
- Dice index for "overall": 93.1653%
32+
- Dice index for "necrotic core / non-enhancing tumor": 77.1611%
33+
- Dice index for "edema": 85.3434%
34+
- Dice index for "enhancing tumor": 84.5571%
35+
2236
See [the original repository](https://github.com/lachinov/brats2018-graphlabunn).
2337

2438

2539
## Performance
2640

2741
## Input
2842

29-
The model takes as an input four MRI modalities `T1`, `T2`, `T1ce`, `Flair`. The inputs are cropped, resamped and z-score normalized. You can find additional information on the BraTS 2018 [page](https://www.med.upenn.edu/sbia/brats2018/data.html) and [wiki](https://en.wikipedia.org/wiki/Magnetic_resonance_imaging).
43+
The model takes as an input four MRI modalities `T1`, `T2`, `T1ce`, `Flair`. The inputs are cropped, resamped and z-score normalized. You can find additional information on the BraTS 2018 [page](https://www.med.upenn.edu/sbia/brats2018/data.html) and [wiki](https://en.wikipedia.org/wiki/Magnetic_resonance_imaging).
3044
In the preprocessing pipeline, all non-zero voxels are cropped and resampled to `128,128,128` resolution first. Then, each modality is z-score normalized separately. The input tensor is a concatenation of the four input modalities.
3145

3246
### Original model

‎models/public/brain-tumor-segmentation-0002/brain-tumor-segmentation-0002.md

+13-13
Original file line numberDiff line numberDiff line change
@@ -17,8 +17,8 @@ This model was created for participation in the [Brain Tumor Segmentation Challe
1717

1818
## Accuracy
1919

20-
See [BRATS 2019 Leaderboard](https://www.cbica.upenn.edu/BraTS19/lboardValidation.html). The metrics
21-
for challenge validation (Dice_WT, Dice_TC, Dice_ET) differ from the metrics reported below (which
20+
See [BRATS 2019 Leaderboard](https://www.cbica.upenn.edu/BraTS19/lboardValidation.html). The metrics
21+
for challenge validation (Dice_WT, Dice_TC, Dice_ET) differ from the metrics reported below (which
2222
are compartible with input labels):
2323

2424
- WT (whole tumor) class combines all three tumor classes:
@@ -30,21 +30,21 @@ are compartible with input labels):
3030
- non-enhancing tumor
3131
- ET (enhancing tumor)
3232

33-
The following accuracy metrics are measured on a `brain tumor` training subset of the [Medical Decathlon](http://medicaldecathlon.com/) dataset.
33+
The following accuracy metrics are measured on a `brain tumor` training subset of the [Medical Decathlon](http://medicaldecathlon.com/) dataset.
3434

3535
**Mean**:
36-
- Dice index for "overall": 0.915
37-
- Dice index for "necrotic core / non-enhancing tumor": 0.806
38-
- Dice index for "edema": 0.611
39-
- Dice index for "enhancing tumor": 0.794
36+
- Dice index for "overall": 91.5%
37+
- Dice index for "necrotic core / non-enhancing tumor": 61.1%
38+
- Dice index for "edema": 80.6%
39+
- Dice index for "enhancing tumor": 79.4%
4040

4141
**Median**:
42-
- Dice index for "overall": 0.927
43-
- Dice index for "necrotic core / non-enhancing tumor": 0.835
44-
- Dice index for "edema": 0.644
45-
- Dice index for "enhancing tumor": 0.86
42+
- Dice index for "overall": 92.7%
43+
- Dice index for "necrotic core / non-enhancing tumor": 64.5%
44+
- Dice index for "edema": 83.5%
45+
- Dice index for "enhancing tumor": 86%
46+
4647

47-
4848
> **NOTE**: The accuracy achieved with ONNX\* model adapted for OpenVINO™ can slightly differ from the accuracy achieved with the original PyTorch model since the upsampling operation was changed from the `trilinear` to `nearest` mode.
4949
5050
## Performance
@@ -131,4 +131,4 @@ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
131131
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
132132
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
133133
THE SOFTWARE.
134-
```
134+
```

‎models/public/caffenet/caffenet.md

+7
Original file line numberDiff line numberDiff line change
@@ -15,6 +15,13 @@ CaffeNet\* model is used for classification. For details see [paper](https://arx
1515
| MParams | 60.965 |
1616
| Source framework | Caffe\* |
1717

18+
## Accuracy
19+
20+
| Metric | Value |
21+
| ------ | ----- |
22+
| Top 1 | 56.714%|
23+
| Top 5 | 79.916%|
24+
1825
## Performance
1926

2027
## Input

‎models/public/ctdet_coco_dlav0_384/ctdet_coco_dlav0_384.md

+1-1
Original file line numberDiff line numberDiff line change
@@ -45,7 +45,7 @@ python convert.py ctdet --load_model /path/to/downloaded/weights.pth --exp_id co
4545

4646
| Metric | Original model | Converted model |
4747
| ------ | -------------- | --------------- |
48-
| MAP | 41.81 | 41.5 |
48+
| mAP | 41.81% | 41.61% |
4949

5050
## Performance
5151

‎models/public/ctdet_coco_dlav0_512/ctdet_coco_dlav0_512.md

+1-1
Original file line numberDiff line numberDiff line change
@@ -45,7 +45,7 @@ python convert.py ctdet --load_model /path/to/downloaded/weights.pth --exp_id co
4545

4646
| Metric | Original model | Converted model |
4747
| ------ | -------------- | --------------- |
48-
| MAP | 44.2 | 44.2 |
48+
| mAP | 44.2% | 44.28% |
4949

5050
## Performance
5151

‎models/public/ctpn/ctpn.md

+6
Original file line numberDiff line numberDiff line change
@@ -15,6 +15,12 @@ Detecting Text in Natural Image with Connectionist Text Proposal Network. For de
1515
| MParams | 17.237 |
1616
| Source framework | TensorFlow\* |
1717

18+
## Accuracy
19+
20+
| Metric | Value |
21+
| ------ | ----- |
22+
| hmean | 1.0864%|
23+
1824
## Performance
1925

2026
## Input

‎models/public/deeplabv3/deeplabv3.md

+4
Original file line numberDiff line numberDiff line change
@@ -17,6 +17,10 @@ DeepLab is a state-of-art deep learning model for semantic image segmentation. F
1717

1818
## Accuracy
1919

20+
| Metric | Value |
21+
| ------ | ----- |
22+
| mean_iou | 66.85%|
23+
2024
## Performance
2125

2226
## Input

‎models/public/densenet-121-caffe2/densenet-121-caffe2.md

+7-2
Original file line numberDiff line numberDiff line change
@@ -4,7 +4,7 @@
44

55
This is a Caffe2\* version of `densenet-121` model, one of the DenseNet
66
group of models designed to perform image classification. This model
7-
was converted from Caffe\* to Caffe2\* format.
7+
was converted from Caffe\* to Caffe2\* format.
88
For details see repository <https://github.com/caffe2/models/tree/master/densenet121>,
99
paper <https://arxiv.org/abs/1608.06993>.
1010

@@ -21,6 +21,11 @@ paper <https://arxiv.org/abs/1608.06993>.
2121

2222
## Accuracy
2323

24+
| Metric | Value |
25+
| ------ | ----- |
26+
| Top 1 | 74.904% |
27+
| Top 5 | 92.192% |
28+
2429
## Performance
2530

2631
## Input
@@ -34,7 +39,7 @@ Image, name - `data`, shape - `1,3,224,224`, format is `B,C,H,W` where:
3439
- `H` - height
3540
- `W` - width
3641

37-
Channel order is `BGR`.
42+
Channel order is `BGR`.
3843
Mean values - [103.94,116.78,123.68], scale value - 58.8235294.
3944

4045
### Converted model

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

+7
Original file line numberDiff line numberDiff line change
@@ -16,6 +16,13 @@ group of models designed to perform image classification. The weights were conve
1616
| MParams | 7.971 |
1717
| Source framework | TensorFlow\* |
1818

19+
## Accuracy
20+
21+
| Metric | Value |
22+
| ------ | ----- |
23+
| Top 1 | 74.29% |
24+
| Top 5 | 91.98%|
25+
1926
## Performance
2027

2128
## Input

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

+9-4
Original file line numberDiff line numberDiff line change
@@ -3,7 +3,7 @@
33
## Use Case and High-Level Description
44

55
The `densenet-121` model is one of the [DenseNet*](https://arxiv.org/abs/1608.06993)
6-
group of models designed to perform image classification. The authors originally trained the models
6+
group of models designed to perform image classification. The authors originally trained the models
77
on Torch\*, but then converted them into Caffe\* format. All DenseNet models have
88
been pretrained on the ImageNet image database. For details about this family of
99
models, check out the [repository](https://github.com/shicai/DenseNet-Caffe).
@@ -22,14 +22,19 @@ models, check out the [repository](https://github.com/shicai/DenseNet-Caffe).
2222

2323
## Accuracy
2424

25+
| Metric | Value |
26+
| ------ | ----- |
27+
| Top 1 | 74.42% |
28+
| Top 5 | 92.136%|
29+
2530
See [the original repository](https://github.com/shicai/DenseNet-Caffe).
2631

2732
## Performance
2833

2934
## Input
3035

3136
The model input is a blob that consists of a single image of 1x3x224x224 in BGR
32-
order. Before passing the image blob into the network, subtract BGR mean values
37+
order. Before passing the image blob into the network, subtract BGR mean values
3338
as follows: [103.94, 116.78, 123.68]. In addition, values must be divided by 0.017.
3439

3540
### Original Model
@@ -41,7 +46,7 @@ Image, name - `data`, shape - `1,3,224,224`, format is `B,C,H,W` where:
4146
- `H` - height
4247
- `W` - width
4348

44-
Channel order is `BGR`.
49+
Channel order is `BGR`.
4550
Mean values - [103.94,116.78,123.68], scale value - 58.8235294117647
4651

4752
### Converted Model
@@ -57,7 +62,7 @@ Channel order is `BGR`.
5762

5863
## Output
5964

60-
The model output for `densenet-121` is a typical object classifier output for 1000 different
65+
The model output for `densenet-121` is a typical object classifier output for 1000 different
6166
classifications matching those in the ImageNet database.
6267

6368
### Original Model

‎models/public/densenet-161-tf/densenet-161-tf.md

+7
Original file line numberDiff line numberDiff line change
@@ -16,6 +16,13 @@ group of models designed to perform image classification. The weights were conve
1616
| MParams | 28.666 |
1717
| Source framework | TensorFlow\* |
1818

19+
## Accuracy
20+
21+
| Metric | Value |
22+
| ------ | ----- |
23+
| Top 1 | 76.446% |
24+
| Top 5 | 93.228%|
25+
1926
## Performance
2027

2128
## Input

‎models/public/densenet-161/densenet-161.md

+6-1
Original file line numberDiff line numberDiff line change
@@ -31,6 +31,11 @@ the 1000 different classifications matching those in the ImageNet database.
3131

3232
## Accuracy
3333

34+
| Metric | Value |
35+
| ------ | ----- |
36+
| Top 1 | 77.55% |
37+
| Top 5 | 93.92% |
38+
3439
See [the original repository](https://github.com/shicai/DenseNet-Caffe).
3540

3641
## Performance
@@ -46,7 +51,7 @@ Image, name - `data`, shape - `1,3,224,224`, format is `B,C,H,W` where:
4651
- `H` - height
4752
- `W` - width
4853

49-
Channel order is `BGR`.
54+
Channel order is `BGR`.
5055
Mean values - [103.94,116.78,123.68], scale value - 58.8235294117647
5156

5257
### Converted model

‎models/public/densenet-169-tf/densenet-169-tf.md

+7
Original file line numberDiff line numberDiff line change
@@ -16,6 +16,13 @@ group of models designed to perform image classification. The weights were conve
1616
| MParams | 14.139 |
1717
| Source framework | TensorFlow\* |
1818

19+
## Accuracy
20+
21+
| Metric | Value |
22+
| ------ | ----- |
23+
| Top 1 | 75.76% |
24+
| Top 5 | 92.81%|
25+
1926
## Performance
2027

2128
## Input

‎models/public/densenet-169/densenet-169.md

+5
Original file line numberDiff line numberDiff line change
@@ -31,6 +31,11 @@ the 1000 different classifications matching those in the ImageNet database.
3131

3232
## Accuracy
3333

34+
| Metric | Value |
35+
| ------ | ----- |
36+
| Top 1 | 76.106% |
37+
| Top 5 | 93.106% |
38+
3439
See [the original repository](https://github.com/shicai/DenseNet-Caffe).
3540

3641
## Performance

‎models/public/densenet-201/densenet-201.md

+5
Original file line numberDiff line numberDiff line change
@@ -31,6 +31,11 @@ the 1000 different classifications matching those in the ImageNet database.
3131

3232
## Accuracy
3333

34+
| Metric | Value |
35+
| ------ | ----- |
36+
| Top 1 | 76.886%|
37+
| Top 5 | 93.556%|
38+
3439
See [the original repository](https://github.com/shicai/DenseNet-Caffe).
3540

3641
## Performance

‎models/public/efficientnet-b0-pytorch/efficientnet-b0-pytorch.md

+2-2
Original file line numberDiff line numberDiff line change
@@ -27,8 +27,8 @@ The model output for `efficientnet-b0-pytorch` is the typical object classifier
2727

2828
| Metric | Original model | Converted model |
2929
| ------ | -------------- | --------------- |
30-
| Top 1 | 76.91 | 76.91 |
31-
| Top 5 | 93.21 | 93.21 |
30+
| Top 1 | 76.91% | 76.91% |
31+
| Top 5 | 93.21% | 93.21% |
3232

3333
## Performance
3434

‎models/public/efficientnet-b0/efficientnet-b0.md

+3-3
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@
22

33
## Use Case and High-Level Description
44

5-
The `efficientnet-b0` model is one of the [EfficientNet](https://arxiv.org/abs/1905.11946) models
5+
The `efficientnet-b0` model is one of the [EfficientNet](https://arxiv.org/abs/1905.11946) models
66
designed to perform image classification.
77
This model was pretrained in TensorFlow\*.
88
All the EfficientNet models have been pretrained on the ImageNet\* image database.
@@ -23,8 +23,8 @@ For details about this family of models, check out the [TensorFlow Cloud TPU rep
2323

2424
| Metric | Original model | Converted model |
2525
| ------ | -------------- | --------------- |
26-
| Top 1 | 75.70 | 75.70 |
27-
| Top 5 | 92.76 | 92.76 |
26+
| Top 1 | 75.70% | 75.70% |
27+
| Top 5 | 92.76% | 92.76% |
2828

2929
## Performance
3030

‎models/public/efficientnet-b0_auto_aug/efficientnet-b0_auto_aug.md

+2-2
Original file line numberDiff line numberDiff line change
@@ -24,8 +24,8 @@ For details about this family of models, check out the [TensorFlow Cloud TPU rep
2424

2525
| Metric | Original model | Converted model |
2626
| ------ | -------------- | --------------- |
27-
| Top 1 | 76.43 | 76.43 |
28-
| Top 5 | 93.04 | 93.04 |
27+
| Top 1 | 76.43% | 76.43% |
28+
| Top 5 | 93.04% | 93.04% |
2929

3030
## Performance
3131

‎models/public/efficientnet-b5-pytorch/efficientnet-b5-pytorch.md

+2-2
Original file line numberDiff line numberDiff line change
@@ -29,8 +29,8 @@ the 1000 different classifications matching those in the ImageNet database.
2929

3030
| Metric | Original model | Converted model |
3131
| ------ | -------------- | --------------- |
32-
| Top 1 | 83.69 | 83.69 |
33-
| Top 5 | 96.71 | 96.71 |
32+
| Top 1 | 83.69% | 83.69% |
33+
| Top 5 | 96.71% | 96.71% |
3434

3535
## Performance
3636

0 commit comments

Comments
 (0)