ssd-mobilenet-v1-variant-1 tensors

submission-guideline.md

@@ -0,0 +1,81 @@
+# General
+The general idea of this tensor-id registry is to provide an identifier for a 
+tensor encoding that is not associated with any specific framework or 
+programming language. The tensor description should be complet enough to allow
+the interpretation of the tensor without additional reference. This register 
+is meant to facilitate N.N. usage in a framework agnostic way. Internal details
+and layers of the N.N. are not necessarily in scope, but if you believe their is value
+in using this register to describe internal layer you are welcome to submit by 
+accompanying your submission with and explanation of the value added. Tensor 
+encoding submission should include at least one reference to the original paper
+from which this encoding emerge or derive, one link to model using this encoding
+ and one link to a code, publically available, able to interpret the tensor encoding.
+
+# Tensor Id
+The identifier of the tensor must be unique for a complet definition of a tensor.
+
+# Tensor Description File
+- Use markdown
+- At the time of writing this, the filename of the file used to described the tensor
+is used as a storage of the tensor-id.
+- Filename is composed of {tensor-id}.md
+
+## Selecting a tensor-id
+
+Anatomy of an tensor-id
+
+1. Only use alpha-numeric character separated by `-` 
+2. Name of the first architecture where this tensor encoding has emerged
+3. Append "-variant-1" term if this encoding is a derivative of the original tensor encoding. If a "variant-1" already exist, use the next variant index. Ex."variant-2". Append "-out" it the tensor is an output of the inference. Use "-in" it is an input of the inference. Use "-inout" if it's output and an input of the inference. 
+4. If the this tensor is part of a tensor-group, append a term that identify the function of this tensor in the tensor-group.
+
+Example:
+
+Original N.N. Architecture: "ssd-mobilenet-v1-variant-1
+
+# Keywords
+
+## Sections
+
+### Internal References
+Reference internal to the registry. These section contain an association 
+between a term used to represent a tensor that is described in a different file.
+Additionnaly if contain a link to the file containing the complete descripiton 
+of the tensor.
+
+### Classification
+
+Classification of the tensor-encoding and tensor-groups.
+
+#### tensor-group
+Indicate if the content describe a tensor-group or a tensor.
+
+
+#### layer-type
+Indicate if the content describe `input`, `output` or `inout` tensor.
+
+#### use-case
+In which context this tensor-encoding is used.
+
+#### part-of-tensor-groups
+List tensor-groups using this tensor, if it is not a generic tensor.
+
+### Description
+Provide a descrition sufficient to understand the information it contain in
+in relation to analysis result and if it's associated to other tensor.
+
+#### tensor-shape
+Provide the tensor dimensions
+
+#### tensor-datatype
+Provide the datatype used by this tensor
+
+### Known Aliases
+List known aliases of this tensor encoding
+
+### Encoding
+
+Describe any pattern used to encode analysis result in the tensor.
+Provide a memory layout of the tensor encoding.
+
+

tensor-groups/ssd-mobilenet-v1-variant-1-out.md

@@ -0,0 +1,53 @@
+# Classification
+- tensor-group: yes
+- layer-type: output
+- use-case: object-detection
+
+# Description
+
+Variant of ssd-mobilenet-v1 outputs. This tensor group is contain of four 
+tensors that together describe the predictions ssd-mobilenet-v1. Count tensor is 
+a scalar the represent the number of object detection in the other 3 tensors.
+
+SSD-MobileNetV1 Variant-1 Output Tensors:
+
+| Name          |shape               |
+|---            |---                 |
+| [count]       |1 x 1               |
+| [boxes]       |1 x [count] x 4     |
+| [classes]     |1 x [count]         |
+| [scores]      |1 x [count]         |
+
+# Tensor Decoding Logic
+
+```
+Foreach i in count:
+    X = boxes[i, 0]
+    Y = boxes[i, 1]
+    W = boxes[i, 2]
+    H = boxes[i, 3]
+    C = classes[i]
+    S = scores[i]
+    detections[i] = [X, Y, W, H, C, S]
+```
+
+# External References
+
+* [MobileNets Paper](https://arxiv.org/pdf/1704.04861)
+* [ONNX model trained on COCO dataset](https://gitlab.collabora.com/gstreamer/onnx-models/-/blob/acc119dd795be5e8c756457dc04507a5d9b8e768/models/ssd_mobilenet_v1_coco.onnx)
+
+# Models
+
+* [ONNX model trained on COCO dataset](https://gitlab.collabora.com/gstreamer/onnx-models/-/blob/acc119dd795be5e8c756457dc04507a5d9b8e768/models/ssd_mobilenet_v1_coco.onnx)
+
+# Tensor Decoders
+|Framework | Links |
+|---       |---    |
+|GStreamer | [perm](https://gitlab.freedesktop.org/gstreamer/gstreamer/-/blob/c206ddd9308a3ce529e0d8957b7c165b3a15c932/subprojects/gst-plugins-bad/gst/tensordecoders/gstssdobjectdetector.c#L36-39) \| [latest](https://gitlab.freedesktop.org/gstreamer/gstreamer/-/blob/main/subprojects/gst-plugins-bad/gst/tensordecoders/gstssdobjectdetector.c?ref_type=heads#L36-39) |
+
+
+[count]: /tensor-groups/generic-variant-1-out-count.md
+[boxes]: /tensors/ssd-mobilenet-v1-variant-1-out-boxes.md
+[classes]: /tensors/ssd-mobilenet-v1-variant-1-out-classes.md
+[scores]: /tensors/ssd-mobilenet-v1-variant-1-out-scores.md
+

tensor-id-register.md

@@ -0,0 +1,8 @@
+# Tensor Id Registry
+
+|Id | Tensor Description Link |
+|---|---                 |
+|generic-variant-1-out-count | [details](/tensors/generic-variant-1-out-count.md) |
+|ssd-mobilenet-v1-variant-1-out-boxes | [details](/tensors/ssd-mobilenet-v1-variant-1-out-boxes.md) |
+|ssd-mobilenet-v1-variant-1-out-classes | [details](/tensors/ssd-mobilenet-v1-variant-1-out-classes.md) |
+|ssd-mobilenet-v1-variant-1-out-scores | [details](/tensors/ssd-mobilenet-v1-variant-1-out-scores.md) |

tensors/generic-variant-1-out-count.md

@@ -0,0 +1,30 @@
+# Classification
+
+- tensor-group: no
+- use-cases
+    - generic
+
+# Description
+
+Scalar expressing a count.
+
+## Count Tensor
+
+- tensor-shape: 1 x 1
+- tensor-datatype: float32, uint8
+- tensor-id: generic-variant-1-count
+
+## Known Aliases
+* num_detecions:0
+
+### Encoding
+Count value stored in the tensor.
+
+Memory layout of tensor data:
+
+|Address| Symbol        |  Value          | Comment                 |
+|---    |---            |---              |---                      |
+| 0x0   | tensor-start  |  count          | complete tensor data    | 
+| 0x4   | tensor-ended  |  -              | -                       | 
+
+

tensors/ssd-mobilenet-v1-variant-1-out-boxes.md

@@ -0,0 +1,63 @@
+# Classification
+
+- tensor-group: no
+- layer-type: output
+- use-case: object-detection
+- part-of-tensor-groups:
+    - [ssd-mobilenet-v1-variant-1-out](/tensor-groups/ssd-mobilenet-v1-variant-1-out.md)
+
+# Description
+Location of objects detected
+
+## Boxes Tensor
+
+- tensor-shape: 1 x [count] x 4
+- tensor-datatype: float32, uint8
+
+### Known Aliases
+* detection_boxes:0
+
+### Encoding
+
+Scheme: (X: box x coord, Y: box y coord, W: box width, H: box height)
+
+|box-1 | box-1 | box-1 | box-1 | box-2 | box-2 | box-2 | box-2 | ... | [count]|[count]|[count]|[count]|
+|---   |---    |---    |---    |---    |---    |---    |---    |---  |---                |---                |---                |---                |
+| X | Y | W | H | X | Y | W | H |...| X | Y | W | H |
+
+Memory layout of tensor data:
+
+|Index                  | Symbol            |Value              | Comment                             |
+|---                    |---                |---                |---                                  |
+| -                     | -                 | -                 | -                                   |
+|0                      |                   |x-coord            |tensor-start, box-0-tensor-data      |
+|1                      |                   |y-coord            |tensor-continue, box-0-tensor-data   |
+|2                      |                   |width              |tensor-continue, box-0-tensor-data   |
+|3                      |                   |height             |tensor-continue, box-0-tensor-data   |
+|4                      |                   |x-coord            |tensor-continue, box-1-tensor-data   |
+|5                      |                   |y-coord            |tensor-continue, box-1-tensor-data   |
+|6                      |                   |width              |tensor-continue, box-1-tensor-data   |
+|7                      |                   |height             |tensor-continue, box-1-tensor-data   |
+|...                    | ...               | ...               | ...                                 |
+|([count] - 1) x 4      |                   |x-coord            |tensor-continue, [count]-tensor-data |
+|([count] - 1) x 4 + 1  |                   |y-coord            |tensor-continue, [count]-tensor-data |
+|([count] - 1) x 4 + 2  |                   |width              |tensor-continue, [count]-tensor-data |
+|([count] - 1) x 4 + 3  |                   |height             |tensor-continue, [count]-tensor-data |
+| -                     | -                 | -                 | -                                   |
+
+# External References
+
+* [MobileNets Paper](https://arxiv.org/pdf/1704.04861)
+* [ONNX model trained on COCO dataset](https://gitlab.collabora.com/gstreamer/onnx-models/-/blob/acc119dd795be5e8c756457dc04507a5d9b8e768/models/ssd_mobilenet_v1_coco.onnx)
+
+# Models
+
+* [ONNX model trained on COCO dataset](https://gitlab.collabora.com/gstreamer/onnx-models/-/blob/acc119dd795be5e8c756457dc04507a5d9b8e768/models/ssd_mobilenet_v1_coco.onnx)
+
+# Tensor Decoders
+|Framework | Links |
+|---       |---    |
+|GStreamer | [perm](https://gitlab.freedesktop.org/gstreamer/gstreamer/-/blob/c206ddd9308a3ce529e0d8957b7c165b3a15c932/subprojects/gst-plugins-bad/gst/tensordecoders/gstssdobjectdetector.c#L36-39) \| [latest](https://gitlab.freedesktop.org/gstreamer/gstreamer/-/blob/main/subprojects/gst-plugins-bad/gst/tensordecoders/gstssdobjectdetector.c?ref_type=heads#L36-39) |
+
+
+[count]: /tensors/generic-variant-1-out-count.md

tensors/ssd-mobilenet-v1-variant-1-out-classes.md

@@ -0,0 +1,58 @@
+# Classification
+
+- tensor-group: no
+- layer-type: output
+- use-case: object-detection
+- part-of-tensor-groups:
+    - [ssd-mobilenet-v1-variant-1-out](/tensor-groups/ssd-mobilenet-v1-variant-1-out.md)
+
+# Description
+
+Classes of object associated with [object location](/tensors/ssd-mobilenet-v1-variant-1-out-boxes.md)
+
+## Classes Tensor
+
+tensor-shape: 1 x [count]
+tensor-datatype: float32, uint8
+
+### Known Aliases
+* detection_classes:0
+
+### Encoding
+
+Scheme:
+|det-1          | det-2         | ... | det-[count]  |
+|---            |---            |---  |---              |
+| object-class  | object-class  | ... | object-class    |
+
+
+Memory layout of tensor data:
+
+|Index         |Value                      | Comment                        |
+|---             |---                        |---                             |
+| -              | -                         | -                              |
+|0               | det-1-objet-class         | tensor-start, tensor-data      |
+|1               | det-2-objet-class         | tensor-continue, tensor-data   |
+|...             | ...                       | tensor-continue, tensor-data   |
+|[count] - 1     | det-[count]-objet-class   | tneosr-continue, tensor-data   | 
+|([count])       | -                         | -                              |
+
+# External References
+
+* [MobileNets Paper](https://arxiv.org/pdf/1704.04861)
+* [ONNX model trained on COCO dataset](https://gitlab.collabora.com/gstreamer/onnx-models/-/blob/acc119dd795be5e8c756457dc04507a5d9b8e768/models/ssd_mobilenet_v1_coco.onnx)
+
+# Models
+
+* [ONNX model trained on COCO dataset](https://gitlab.collabora.com/gstreamer/onnx-models/-/blob/acc119dd795be5e8c756457dc04507a5d9b8e768/models/ssd_mobilenet_v1_coco.onnx)
+
+# Tensor Decoders
+|Framework | Links |
+|---       |---    |
+|GStreamer | [perm](https://gitlab.freedesktop.org/gstreamer/gstreamer/-/blob/c206ddd9308a3ce529e0d8957b7c165b3a15c932/subprojects/gst-plugins-bad/gst/tensordecoders/gstssdobjectdetector.c#L36-39) \| [latest](https://gitlab.freedesktop.org/gstreamer/gstreamer/-/blob/main/subprojects/gst-plugins-bad/gst/tensordecoders/gstssdobjectdetector.c?ref_type=heads#L36-39) |
+
+
+
+
+
+

tensors/ssd-mobilenet-v1-variant-1-out-scores.md

@@ -0,0 +1,53 @@
+# Classification
+
+- tensor-group: no
+- layer-type: output
+- use-case: object-detection
+- part-of-tensor-groups:
+    - [ssd-mobilenet-v1-variant-1-out](/tensor-groups/ssd-mobilenet-v1-variant-1-out.md)
+
+# Description
+
+Confidence levels of [classes](/tensors/ssd-mobilenet-v1-variant-1-out-classes.md) associated [object location](/tensors/ssd-mobilenet-v1-variant-1-out-boxes.md)
+
+## Scores Tensor
+
+tensor-shape: 1 x [count]
+tensor-datatype: float32, uint8
+tensor-id: out-ssd-mobilenet-v1-scores-variant-1
+
+### Known Aliases
+* detection_scores:0
+
+### Encoding
+Number of detection correspond to the value stored in the tensor.
+
+Memory layout of tensor data:
+
+|Index                            | Symbol                          |Value                                | Comment                 |
+|---                              |---                              |---                                  |---                      |
+| -                               | -                               | -                                   | -                       |
+|0                                | tensor-start                    | det-1-objet-class-score             | tensor-data, start      |
+|1 i                              | tensor-continue                 | det-2-objet-class-score             | tensor-data, continue   |
+|...                              | ...                             | ...                                 | tensor-data, continue   |
+|([count] - 1)                    | tensor-end                      | det-[count]-objet-class-score       | tensor-data, end        |
+|[count]                          | -                               | -                                   | -                       |
+
+# External References
+
+* [MobileNets Paper](https://arxiv.org/pdf/1704.04861)
+
+# Models
+
+* [ONNX model trained on COCO dataset](https://gitlab.collabora.com/gstreamer/onnx-models/-/blob/acc119dd795be5e8c756457dc04507a5d9b8e768/models/ssd_mobilenet_v1_coco.onnx)
+
+# Tensor Decoders
+|Framework | Links |
+|---       |---    |
+|GStreamer | [perm](https://gitlab.freedesktop.org/gstreamer/gstreamer/-/blob/c206ddd9308a3ce529e0d8957b7c165b3a15c932/subprojects/gst-plugins-bad/gst/tensordecoders/gstssdobjectdetector.c#L36-39) \| [latest](https://gitlab.freedesktop.org/gstreamer/gstreamer/-/blob/main/subprojects/gst-plugins-bad/gst/tensordecoders/gstssdobjectdetector.c?ref_type=heads#L36-39) |
+
+
+[count]: /tensors/generic-variant-1-out-count.md
+
+
+