efficientdet-d0-tf

Use Case and High-Level Description

The efficientdet-d0-tf model is one of the EfficientDet models designed to perform object detection. This model was pre-trained in TensorFlow*. All the EfficientDet models have been pre-trained on the Common Objects in Context (COCO) image database. For details about this family of models, check out the Google AutoML repository.

Steps to Reproduce Conversion to Frozen Graph

1. Clone the original repository

git clone https://github.com/google/automl.git
cd automl

2. Checkout the commit that the conversion was tested on:

git checkout 341af7d4da7805c3a874877484e133f33c420ec5

3. Navigate to efficientdet source code directory

cd efficientdet

4. Install dependencies

pip install -r requirements.txt

5. Download model checkpoint archive using this link and unzip it.
6. Run following command:

   python model_inspect.py --runmode=saved_model --model_name=efficientdet-d0 --ckpt_path=CHECKPOINT_DIR --saved_model_dir=OUTPUT_DIR
   

   where CHECKPOINT_DIR - directory where model checkpoint stored, OUTPUT_DIR - directory where converted model should be stored.

Specification

Metric	Value
Type	Object detection
GFLOPs	2.54
MParams	3.9
Source framework	TensorFlow*

Accuracy

Metric	Converted model
COCO mAP (0.5:0.05:0.95)	31.95%

Input

Original Model

Image, name - image_arrays, shape - 1, 512, 512, 3, format is B, H, W, C, where:

• B - batch size
• H - height
• W - width
• C - channel

Channel order is RGB.

Converted Model

Image, name - image_arrays/placeholder_port_0, shape - 1, 512, 512, 3, format is B, H, W, C, where:

• B - batch size
• H - height
• W - width
• C - channel

Channel order is BGR.

Output

Original Model

The array of summary detection information, name: detections, shape: 1, 100, 7 in the format 1, N, 7, where N is the number of detected bounding boxes. For each detection, the description has the format: [image_id, y_min, x_min, y_max, x_max, confidence, label], where:

• image_id - ID of the image in the batch
• (x_min, y_min) - coordinates of the top left bounding box corner
• (x_max, y_max) - coordinates of the bottom right bounding box corner
• confidence - confidence for the predicted class
• label - predicted class ID, in range [1, 91], mapping to class names provided in <omz_dir>/data/dataset_classes/coco_91cl.txt file

Converted Model

The array of summary detection information, name: detections, shape: 1, 1, 100, 7 in the format 1, 1, N, 7, where N is the number of detected bounding boxes. For each detection, the description has the format: [image_id, label, conf, x_min, y_min, x_max, y_max], where:

• image_id - ID of the image in the batch
• label - predicted class ID, in range [0, 90], mapping to class names provided in <omz_dir>/data/dataset_classes/coco_91cl.txt file
• conf - confidence for the predicted class
• (x_min, y_min) - coordinates of the top left bounding box corner (coordinates stored in normalized format, in range [0, 1])
• (x_max, y_max) - coordinates of the bottom right bounding box corner (coordinates stored in normalized format, in range [0, 1])

Download a Model and Convert it into OpenVINO™ IR Format

You can download models and if necessary convert them into OpenVINO™ IR format using the Model Downloader and other automation tools as shown in the examples below.

An example of using the Model Downloader:

omz_downloader --name <model_name>

An example of using the Model Converter:

omz_converter --name <model_name>

Demo usage

The model can be used in the following demos provided by the Open Model Zoo to show its capabilities:

• Object Detection C++ Demo
• Object Detection Python* Demo
• Pedestrian Tracker C++ Demo

Legal Information

The original model is distributed under the Apache License, Version 2.0. A copy of the license is provided in <omz_dir>/models/public/licenses/APACHE-2.0-TF-AutoML.txt.