Skip to content

Commit 3bcadb3

Browse files
author
Roman Donchenko
authored
Merge pull request openvinotoolkit#260 from jkamelin/jk/tf_models
TF models for R3
2 parents 6953019 + f322bbe commit 3bcadb3

File tree

9 files changed

+1567
-18
lines changed

9 files changed

+1567
-18
lines changed

‎models/public/mask_rcnn_inception_resnet_v2_atrous_coco/mask_rcnn_inception_resnet_v2_atrous_coco.md

+78
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,78 @@
1+
# mask_rcnn_inception_resnet_v2_atrous_coco
2+
3+
## Use Case and High-Level Description
4+
5+
Mask R-CNN Inception Resnet V2 Atrous trained on COCO dataset. Used for object instance segmentation. For details, see [paper](https://arxiv.org/pdf/1703.06870.pdf).
6+
7+
## Example
8+
9+
## Specification
10+
11+
| Metric | Value |
12+
|---------------------------------|-------------------------------------------|
13+
| Type | Instance segmentation |
14+
| GFlops | 675.314 |
15+
| MParams | 92.368 |
16+
| Source framework | TensorFlow\* |
17+
18+
## Performance
19+
20+
## Input
21+
22+
### Original Model
23+
24+
Image, name: `image_tensor`, shape: [1x800x800x3], format: [BxHxWxC],
25+
where:
26+
27+
- B - batch size
28+
- H - image height
29+
- W - image width
30+
- C - number of channels
31+
32+
Expected color order: RGB.
33+
34+
### Converted Model
35+
36+
1. Image, name: `image_tensor`, shape: [1x3x800x800], format: [BxCxHxW],
37+
where:
38+
39+
- B - batch size
40+
- C - number of channels
41+
- H - image height
42+
- W - image width
43+
44+
Expected color order: BGR.
45+
46+
2. Information of input image size, name: `image_info`, shape: [1x3], format: [BxC],
47+
where:
48+
49+
- B - batch size
50+
- C - vector of 3 values in format [H,W,S], where H is an image height, W is an image width, S is an image scale factor (usually 1)
51+
52+
## Output
53+
54+
### Original Model
55+
56+
1. Classifier, name: `detection_classes`. Contains predicted bounding boxes classes in a range [1, 91]. The model was trained on the Microsoft\* COCO dataset version with 90 categories of objects, 0 class is for background.
57+
2. Probability, name: `detection_scores`. Contains probability of detected bounding boxes.
58+
3. Detection box, name: `detection_boxes`. Contains detection boxes coordinates in a format `[y_min, x_min, y_max, x_max]`, where (`x_min`, `y_min`) are coordinates of the top left corner, (`x_max`, `y_max`) are coordinates of the right bottom corner. Coordinates are rescaled to input image size.
59+
4. Detections number, name: `num_detections`. Contains the number of predicted detection boxes.
60+
5. Segmentation mask, name: `detection_masks`. Contains segmentation heatmaps of detected objects for all classes for every output bounding box.
61+
62+
### Converted Model
63+
64+
1. The array of summary detection information, name: `reshape_do_2d`, shape: [N, 7], where N is the number of detected
65+
bounding boxes. For each detection, the description has the format:
66+
[`image_id`, `label`, `conf`, `x_min`, `y_min`, `x_max`, `y_max`],
67+
where:
68+
69+
- `image_id` - ID of the image in the batch
70+
- `label` - predicted class ID
71+
- `conf` - confidence for the predicted class
72+
- (`x_min`, `y_min`) - coordinates of the top left bounding box corner (coordinates stored in normalized format, in range [0, 1])
73+
- (`x_max`, `y_max`) - coordinates of the bottom right bounding box corner (coordinates stored in normalized format, in range [0, 1])
74+
2. Segmentation heatmaps for all classes for every output bounding box, name: `masks`, shape: [N, 90, 33, 33], where N is the number of detected masks, 90 is the number of classes, with background class excluded.
75+
76+
## Legal Information
77+
78+
[https://raw.githubusercontent.com/tensorflow/models/master/LICENSE]()

‎models/public/mask_rcnn_inception_v2_coco/mask_rcnn_inception_v2_coco.md

+78
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,78 @@
1+
# mask_rcnn_inception_v2_coco
2+
3+
## Use Case and High-Level Description
4+
5+
Mask R-CNN Inception V2 trained on COCO dataset. Used for object instance segmentation. For details, see [paper](https://arxiv.org/pdf/1703.06870.pdf).
6+
7+
## Example
8+
9+
## Specification
10+
11+
| Metric | Value |
12+
|---------------------------------|-------------------------------------------|
13+
| Type | Instance segmentation |
14+
| GFlops | 54.926 |
15+
| MParams | 21.772 |
16+
| Source framework | TensorFlow\* |
17+
18+
## Performance
19+
20+
## Input
21+
22+
### Original Model
23+
24+
Image, name: `image_tensor`, shape: [1x800x800x3], format: [BxHxWxC],
25+
where:
26+
27+
- B - batch size
28+
- H - image height
29+
- W - image width
30+
- C - number of channels
31+
32+
Expected color order: RGB.
33+
34+
### Converted Model
35+
36+
1. Image, name: `image_tensor`, shape: [1x3x800x800], format: [BxCxHxW],
37+
where:
38+
39+
- B - batch size
40+
- C - number of channels
41+
- H - image height
42+
- W - image width
43+
44+
Expected color order: BGR.
45+
46+
2. Information of input image size, name: `image_info`, shape: [1x3], format: [BxC],
47+
where:
48+
49+
- B - batch size
50+
- C - vector of 3 values in format [H,W,S], where H is an image height, W is an image width, S is an image scale factor (usually 1)
51+
52+
## Output
53+
54+
### Original Model
55+
56+
1. Classifier, name: `detection_classes`. Contains predicted bounding boxes classes in a range [1, 91]. The model was trained on the Microsoft\* COCO dataset version with 90 categories of objects, 0 class is for background.
57+
2. Probability, name: `detection_scores`. Contains probability of detected bounding boxes.
58+
3. Detection box, name: `detection_boxes`. Contains detection boxes coordinates in a format `[y_min, x_min, y_max, x_max]`, where (`x_min`, `y_min`) are coordinates of the top left corner, (`x_max`, `y_max`) are coordinates of the right bottom corner. Coordinates are rescaled to input image size.
59+
4. Detections number, name: `num_detections`. Contains the number of predicted detection boxes.
60+
5. Segmentation mask, name: `detection_masks`. Contains segmentation heatmaps of detected objects for all classes for every output bounding box.
61+
62+
### Converted Model
63+
64+
1. The array of summary detection information, name: `reshape_do_2d`, shape: [N, 7], where N is the number of detected
65+
bounding boxes. For each detection, the description has the format:
66+
[`image_id`, `label`, `conf`, `x_min`, `y_min`, `x_max`, `y_max`],
67+
where:
68+
69+
- `image_id` - ID of the image in the batch
70+
- `label` - predicted class ID
71+
- `conf` - confidence for the predicted class
72+
- (`x_min`, `y_min`) - coordinates of the top left bounding box corner (coordinates stored in normalized format, in range [0, 1])
73+
- (`x_max`, `y_max`) - coordinates of the bottom right bounding box corner (coordinates stored in normalized format, in range [0, 1])
74+
2. Segmentation heatmaps for all classes for every output bounding box, name: `masks`, shape: [N, 90, 15, 15], where N is the number of detected masks, 90 is the number of classes, with background class excluded.
75+
76+
## Legal Information
77+
78+
[https://raw.githubusercontent.com/tensorflow/models/master/LICENSE]()

‎models/public/mask_rcnn_resnet101_atrous_coco/mask_rcnn_resnet101_atrous_coco.md

+78
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,78 @@
1+
# mask_rcnn_resnet101_atrous_coco
2+
3+
## Use Case and High-Level Description
4+
5+
Mask R-CNN Resnet101 Atrous trained on COCO dataset. Used for object instance segmentation. For details, see [paper](https://arxiv.org/pdf/1703.06870.pdf).
6+
7+
## Example
8+
9+
## Specification
10+
11+
| Metric | Value |
12+
|---------------------------------|-------------------------------------------|
13+
| Type | Instance segmentation |
14+
| GFlops | 674.58 |
15+
| MParams | 69.188 |
16+
| Source framework | TensorFlow\* |
17+
18+
## Performance
19+
20+
## Input
21+
22+
### Original Model
23+
24+
Image, name: `image_tensor`, shape: [1x800x800x3], format: [BxHxWxC],
25+
where:
26+
27+
- B - batch size
28+
- H - image height
29+
- W - image width
30+
- C - number of channels
31+
32+
Expected color order: RGB.
33+
34+
### Converted Model
35+
36+
1. Image, name: `image_tensor`, shape: [1x3x800x800], format: [BxCxHxW],
37+
where:
38+
39+
- B - batch size
40+
- C - number of channels
41+
- H - image height
42+
- W - image width
43+
44+
Expected color order: BGR.
45+
46+
2. Information of input image size, name: `image_info`, shape: [1x3], format: [BxC],
47+
where:
48+
49+
- B - batch size
50+
- C - vector of 3 values in format [H,W,S], where H is an image height, W is an image width, S is an image scale factor (usually 1)
51+
52+
## Output
53+
54+
### Original Model
55+
56+
1. Classifier, name: `detection_classes`. Contains predicted bounding boxes classes in a range [1, 91]. The model was trained on the Microsoft\* COCO dataset version with 90 categories of objects, 0 class is for background.
57+
2. Probability, name: `detection_scores`. Contains probability of detected bounding boxes.
58+
3. Detection box, name: `detection_boxes`. Contains detection boxes coordinates in a format `[y_min, x_min, y_max, x_max]`, where (`x_min`, `y_min`) are coordinates of the top left corner, (`x_max`, `y_max`) are coordinates of the right bottom corner. Coordinates are rescaled to input image size.
59+
4. Detections number, name: `num_detections`. Contains the number of predicted detection boxes.
60+
5. Segmentation mask, name: `detection_masks`. Contains segmentation heatmaps of detected objects for all classes for every output bounding box.
61+
62+
### Converted Model
63+
64+
1. The array of summary detection information, name: `reshape_do_2d`, shape: [N, 7], where N is the number of detected
65+
bounding boxes. For each detection, the description has the format:
66+
[`image_id`, `label`, `conf`, `x_min`, `y_min`, `x_max`, `y_max`],
67+
where:
68+
69+
- `image_id` - ID of the image in the batch
70+
- `label` - predicted class ID
71+
- `conf` - confidence for the predicted class
72+
- (`x_min`, `y_min`) - coordinates of the top left bounding box corner (coordinates stored in normalized format, in range [0, 1])
73+
- (`x_max`, `y_max`) - coordinates of the bottom right bounding box corner (coordinates stored in normalized format, in range [0, 1])
74+
2. Segmentation heatmaps for all classes for every output bounding box, name: `masks`, shape: [N, 90, 33, 33], where N is the number of detected masks, 90 is the number of classes, with background class excluded.
75+
76+
## Legal Information
77+
78+
[https://raw.githubusercontent.com/tensorflow/models/master/LICENSE]()

‎models/public/mask_rcnn_resnet50_atrous_coco/mask_rcnn_resnet50_atrous_coco.md

+78
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,78 @@
1+
# mask_rcnn_resnet50_atrous_coco
2+
3+
## Use Case and High-Level Description
4+
5+
Mask R-CNN Resnet50 Atrous trained on COCO dataset. Used for object instance segmentation. For details, see [paper](https://arxiv.org/pdf/1703.06870.pdf).
6+
7+
## Example
8+
9+
## Specification
10+
11+
| Metric | Value |
12+
|---------------------------------|-------------------------------------------|
13+
| Type | Instance segmentation |
14+
| GFlops | 294.738 |
15+
| MParams | 50.222 |
16+
| Source framework | TensorFlow\* |
17+
18+
## Performance
19+
20+
## Input
21+
22+
### Original Model
23+
24+
Image, name: `image_tensor`, shape: [1x800x800x3], format: [BxHxWxC],
25+
where:
26+
27+
- B - batch size
28+
- H - image height
29+
- W - image width
30+
- C - number of channels
31+
32+
Expected color order: RGB.
33+
34+
### Converted Model
35+
36+
1. Image, name: `image_tensor`, shape: [1x3x800x800], format: [BxCxHxW],
37+
where:
38+
39+
- B - batch size
40+
- C - number of channels
41+
- H - image height
42+
- W - image width
43+
44+
Expected color order: BGR.
45+
46+
2. Information of input image size, name: `image_info`, shape: [1x3], format: [BxC],
47+
where:
48+
49+
- B - batch size
50+
- C - vector of 3 values in format [H,W,S], where H is an image height, W is an image width, S is an image scale factor (usually 1)
51+
52+
## Output
53+
54+
### Original Model
55+
56+
1. Classifier, name: `detection_classes`. Contains predicted bounding boxes classes in a range [1, 91]. The model was trained on the Microsoft\* COCO dataset version with 90 categories of objects, 0 class is for background.
57+
2. Probability, name: `detection_scores`. Contains probability of detected bounding boxes.
58+
3. Detection box, name: `detection_boxes`. Contains detection boxes coordinates in a format `[y_min, x_min, y_max, x_max]`, where (`x_min`, `y_min`) are coordinates of the top left corner, (`x_max`, `y_max`) are coordinates of the right bottom corner. Coordinates are rescaled to input image size.
59+
4. Detections number, name: `num_detections`. Contains the number of predicted detection boxes.
60+
5. Segmentation mask, name: `detection_masks`. Contains segmentation heatmaps of detected objects for all classes for every output bounding box.
61+
62+
### Converted Model
63+
64+
1. The array of summary detection information, name: `reshape_do_2d`, shape: [N, 7], where N is the number of detected
65+
bounding boxes. For each detection, the description has the format:
66+
[`image_id`, `label`, `conf`, `x_min`, `y_min`, `x_max`, `y_max`],
67+
where:
68+
69+
- `image_id` - ID of the image in the batch
70+
- `label` - predicted class ID
71+
- `conf` - confidence for the predicted class
72+
- (`x_min`, `y_min`) - coordinates of the top left bounding box corner (coordinates stored in normalized format, in range [0, 1])
73+
- (`x_max`, `y_max`) - coordinates of the bottom right bounding box corner (coordinates stored in normalized format, in range [0, 1])
74+
2. Segmentation heatmaps for all classes for every output bounding box, name: `masks`, shape: [N, 90, 33, 33], where N is the number of detected masks, 90 is the number of classes, with background class excluded.
75+
76+
## Legal Information
77+
78+
[https://raw.githubusercontent.com/tensorflow/models/master/LICENSE]()

‎models/public/mobilenet-v2-1.0-224/mobilenet-v2-1.0-224.md

+61
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,61 @@
1+
# mobilenet-v2-1.0-224
2+
3+
## Use Case and High-Level Description
4+
5+
`mobilenet-v2-1.0-224` is one of MobileNets - small, low-latency, low-power models parameterized to meet the resource constraints of a variety of use cases. They can be built upon for classification, detection, embeddings and segmentation similar to how other popular large scale models are used. For details, see [paper](https://arxiv.org/abs/1704.04861).
6+
7+
## Example
8+
9+
## Specification
10+
11+
| Metric | Value |
12+
|---------------------------------|-------------------------------------------|
13+
| Type | Classification |
14+
| GFlops | 0.615 |
15+
| MParams | 3.489 |
16+
| Source framework | TensorFlow\* |
17+
18+
## Performance
19+
20+
## Input
21+
22+
### Original Model
23+
24+
Image, name: `input` , shape: [1x224x224x3], format: [BxHxWxC], where:
25+
26+
- B - batch size
27+
- H - image height
28+
- W - image width
29+
- C - number of channels
30+
31+
Expected color order: RGB.
32+
Mean values: [127.5, 127.5, 127.5], scale factor for each channel: 127.0
33+
34+
### Converted Model
35+
36+
Image, name: `input` , shape: [1x3x224x224], format: [BxCxHxW], where:
37+
38+
- B - batch size
39+
- C - number of channels
40+
- H - image height
41+
- W - image width
42+
43+
Expected color order: BGR.
44+
45+
## Output
46+
47+
### Original Model
48+
49+
Probabilities for all dataset classes (0 class is background). Probabilities are represented in logits format. Name: `MobilenetV2/Predictions/Reshape_1`.
50+
51+
### Converted Model
52+
53+
Probabilities for all dataset classes (0 class is background). Probabilities are represented in logits format. Name: `MobilenetV2/Predictions/Softmax`, shape: [1,1001], format: [BxC],
54+
where:
55+
56+
- B - batch size
57+
- C - vector of probabilities.
58+
59+
## Legal Information
60+
61+
[https://raw.githubusercontent.com/tensorflow/models/master/LICENSE]()

0 commit comments

Comments
 (0)