Adding models from Arm internal zoo

Adding efficientnet_lite0_224,
yolov3_416_416_backbone_mltools_int8 and
yolov3_tiny_int8_pruned_backbone_only
from Arm internal zoo to public zoo models/experimental.

models/experimental/efficientnet_lite0_224/README.md

@@ -0,0 +1,37 @@
+# image_classification/efficientnet_lite0_224/tflite_int8
+
+## Description
+This work is developed from the codebase located [here](https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/lite/README.md) and is under an Apache 2 license available [here](https://github.com/tensorflow/tpu/blob/master/LICENSE). 
+
+The original networks, which we have optimized via tooling but left otherwise unchanged are copyright the tensorflow authors as in the license file linked.
+
+## License
+[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)
+
+## Network Information
+| Network Information | Value |
+|---------------------|-------|
+|  Framework          | TensorFlow Lite |
+|  SHA-1 Hash         | 35f9dafaf25f8abf2225265b0724979a68bf6d67 |
+|  Size (Bytes)       | 5422760 |
+|  Provenance         | https://storage.googleapis.com/cloud-tpu-checkpoints/efficientnet/lite/efficientnet-lite0.tar.gz |
+|  Paper | https://arxiv.org/pdf/1905.11946.pdf |
+
+
+## Accuracy
+Dataset: ILSVRC 2012
+
+| Metric | Value |
+|--------|-------|
+| top_1_accuracy | 0.744 |
+
+## Network Inputs
+| Input Node Name | Shape | Example Path | Example Type | Example Use Case |
+|-----------------|-------|--------------|------------------|--------------|
+| images | (1, 224, 224, 3) | models/image_classification/efficientnet_lite0_224/tflite_int8/testing_input |  | Typical ImageNet-style single-batch cat resized to 224x224. |
+
+## Network Outputs
+| Output Node Name | Shape | Description |
+|------------------|-------|-------------|
+| Softmax | (1, 1000) | Probability distribution over 1000 ImageNet classes with uint8 values. |
+

models/experimental/yolov3_416_416_backbone_mltools_int8/README.md

@@ -0,0 +1,38 @@
+# object_detection/yolo_v3_backbone_mltools/tflite_int8
+
+## Description
+Backbone of the Yolo v3 model with an input size of 416 x 416. The backbone is quantized with an int8 precision using the first 1000 images of the COCO 2014 training set for calibration. The DarkNet original pre-trained weights are used as initial weights.
+
+## License
+[MIT](https://spdx.org/licenses/MIT.html)
+[MIT]https://github.com/zzh8829/yolov3-tf2/blob/master/LICENSE
+
+## Network Information
+| Network Information | Value |
+|---------------------|-------|
+|  Framework          | TensorFlow Lite |
+|  SHA-1 Hash         | 4adc0b716c5af29d957396fab2bcbc460e8b94ee |
+|  Size (Bytes)       | 62958128 |
+|  Provenance         | https://confluence.arm.com/display/MLENG/Yolo+v3 |
+|  Paper | https://pjreddie.com/media/files/papers/YOLOv3.pdf |
+
+
+## Accuracy
+Dataset: coco-val-2014
+
+| Metric | Value |
+|--------|-------|
+| mAP50 | 0.563 |
+
+## Network Inputs
+| Input Node Name | Shape | Example Path | Example Type | Example Use Case |
+|-----------------|-------|--------------|------------------|--------------|
+| input_int8 | (1, 416, 416, 3) | models/object_detection/yolo_v3_backbone_mltools/tflite_int8/testing_input/0.npy | int8 |  |
+
+## Network Outputs
+| Output Node Name | Shape | Description |
+|------------------|-------|-------------|
+| Identity_int8 | (1, 13, 13, 3, 85) | None |
+| Identity_1_int8 | (1, 26, 26, 3, 85) | None |
+| Identity_2_int8 | (1, 52, 52, 3, 85) | None |
+

models/experimental/yolov3_tiny_int8_pruned_backbone_only/README.md

@@ -0,0 +1,44 @@
+# object_detection/yolo_v3_tiny/tflite_pruned_backbone_only_int8
+
+## Description
+YOLO v3 Tiny is the light version of YOLO v3 with less layers for object detection and classification.  
+  This model contains only the backbone, and using Darknet pre-trained weights.
+
+## License
+[MIT License](https://github.com/zzh8829/yolov3-tf2/blob/master/LICENSE)
+
+## Network Information
+| Network Information | Value |
+|---------------------|-------|
+|  Framework          | TensorFlow Lite |
+|  SHA-1 Hash         | ec4c5ad5c92fe6bb7eb750011b0b1e322a15ba19 |
+|  Size (Bytes)       | 8963352 |
+|  Provenance         | https://github.com/zzh8829/yolov3-tf2 + https://pjreddie.com/media/files/yolov3-tiny.weights |
+|  Paper | https://arxiv.org/pdf/1804.02767.pdf |
+
+
+## DataSet
+| Dataset Information | Value |
+|--------|-------|
+| Name | Microsoft Coco 2014 |
+| Description | COCO is a large-scale object detection, segmentation, and captioning dataset. |
+| Link | https://cocodataset.org/#home |
+
+
+## Accuracy
+
+| Metric | Value |
+|--------|-------|
+| mAP | 0.345 |
+
+## Network Inputs
+| Input Node Name | Shape | Type | Example Path | Example Type | Example Shape | Example Use Case |
+|-----------------|-------|-------|--------------|-------|-------|-----------------|
+| serving_default_input:0 | (1, 416, 416, 3) | int8 | models/object_detection/yolo_v3_tiny/tflite_pruned_backbone_only_int8/testing_input/serving_default_input:0 | int8 | [1, 416, 416, 3] | Random input for model regression. |
+
+## Network Outputs
+| Output Node Name | Shape | Type | Example Path | Example Type | Example Shape | Example Use Case |
+|-----------------|-------|-------|--------------|-------|-------|-----------------|
+| StatefulPartitionedCall:0 | (1, 13, 13, 3, 85) | int8 | models/object_detection/yolo_v3_tiny/tflite_pruned_backbone_only_int8/testing_output/StatefulPartitionedCall:0 | int8 | [1, 13, 13, 3, 85] | output for model regression. |
+| StatefulPartitionedCall:1 | (1, 26, 26, 3, 85) | int8 | models/object_detection/yolo_v3_tiny/tflite_pruned_backbone_only_int8/testing_output/StatefulPartitionedCall:1 | int8 | [1, 26, 26, 3, 85] | output for model regression. |
+