Datature Code and Resources Hub

A repository of resources used in our tutorials and guides ⚡️

The scripts & jupyter notebooks in this repository provide Nexus users a guideline for those who may want to load a model for prediction and integrate them into their codebase or modify some parameters instead of using Portal to predict directly.

Getting Started

Firstly, users should clone this repository and change to the resource folder directory.

git clone https://github.com/datature/resources.git

Users can download two kinds of model formats from Nexus currently: Tensorflow model and TFlite model.

• Tensorflow Model - Bounding Box (Direct Download)
• Tensorflow Model - Bounding Box (Datature Hub Download)
• Tensorflow Model - Masks (Direct Download)
• TFLite Model - Bounding Box (Direct Download)

Download Model

Users can download their trained model directly from Nexus or port the trained model through Datature Hub. Users need two sets of keys for the second method: Model Key and Project Secret Key.

Model Key

To convert that artifact into an exported model for the prediction service, in Nexus, select Artifacts under Project Overview. Within the artifacts page, select your chosen artifact and model format to generate a model key for deployment by clicking the triple dots box shown below.

Project Secret Key

You can generate the project secret key on the Nexus platform by going to API Management and selecting the Generate New Secret button, as shown below.

Environment Requirements

python 3.7 =< version =<3.9
Jupyter Notebook
tensorflow >= 2.5.0

Tensorflow Models

Click to expand

Under the scripts/inference/tensorflow/ folder, there are subfolders of different types of computer vision models application or configuration, in Tensorflow format:

• Tensorflow Model - Bounding Box
• Tensorflow Model - Masks

File Structure

Each of the subfolders contains a standard file structure. Below are descriptions for each file and folder regarding its content or purpose.

• input/: Some sample test images for prediction
• output/: Output folder to store predicted images
• requirements.txt: Python3 dependencies
• model_architecture/: All bounding box model architectures offered on Nexus platform
  • predict.py: Python script to load downloaded model and obtain predictions
  • tf_xxx.ipynb: Jupyter notebook script to load model and obtain predictions
  • predict_hub.py: Python script to load model directly from Nexus platform

Tensorflow Model - Bounding Box

Click to expand

xxx in the subsequent command prompts represents chosen model architecture, which is also the name of the folders within the model_architecture/ folder.

Command to run Downloaded Model Python script

cd scripts/inference/tensorflow/bounding_box/

pip install -r requirements.txt

cd model_architecture/xxx

python predict.py --input "path_to_input_folder" --output "path_to_output_folder" --width 640 --height 640 --threshold 0.7 --model "path_to_model" --label "path_to_labelmap"

Sample default command

python predict.py --input "./input" --output "./output" --width 640 --height 640 --threshold 0.7 --model "./saved_model" --label "./label_map.pbtxt"

Below is the list of modifiable script parameters and its description.

--input "path_to_input_folder" (Required)
--output "path_to_output_folder" (Required)
--model "path_to_model" (Required)
--label "path_to_labelmap" (Required)
--width "width of image to load" (Optional) (default: 640)
--height "height of image to load" (Optional) (default: 640)
--threshold "confidence threshold" (Optional) (default: 0.7)

Command to run Jupyter Notebook

pip install jupyter

python -m notebook tf_xxx.ipynb

Command to run Datature Hub Python script

cd scripts/inference/tensorflow/bounding_box/

pip install -r requirements.txt

cd model_architecture/xxx

python predict_hub.py --input "path_to_input_folder" --output "path_to_output_folder"  --threshold 0.7 --secret "Project_secret" --key "Your_model_key"

Sample default command

python predict_hub.py --input "./input" --output "./output" --secret "76d97105923491bfa13c84d74eb5457b3b04dceda19ca009d7af111bd7d05344" --key "f2324a0064025c01da8fe3482177a83a"

Below is the list of modifiable script parameters and its description.

--input "Path to folder that contains input images" (Required) (default:"./input")
--output "Path to folder to store predicted images" (Required)(default:"./output")
--threshold "Prediction confidence threshold" (Optional) (default: 0.7)
--secret "Datature Nexus project secret key" (Required)
--key "Datature Nexus model key" (Required)

↥ back to top

Tensorflow Model - Masks

Click to expand

xxx in the subsequent command prompts represents chosen model architecture, which is also the name of the folders within the model_architecture/ folder.

Command to run Downloaded Model Python script

cd scripts/inference/tensorflow/tensorflow/segmentation/

pip install -r requirements.txt

cd model_architecture/xxx

python predict.py --input "path_to_input_folder" --output "path_to_output_folder" --width 1024 --height 1024 --threshold 0.7 --model "path_to_model" --label "path_to_labelmap"

Sample default command

python predict.py --input "./input" --output "./output" --width 1024 --height 1024 --threshold 0.7 --model "./saved_model" --label "./label_map.pbtxt"

Below is the list of modifiable script parameters and its description.

--input "path_to_input_folder" (Required)
--output "path_to_output_folder" (Required)
--model "path_to_model" (Required)
--label "path_to_labelmap" (Required)
--width "width of image to load" (Optional) (default: 1024)
--height "height of image to load" (Optional) (default: 1024)
--threshold "confidence threshold" (Optional) (default: 0.7)

Command to run Jupyter Notebook

pip install jupyter

python -m notebook tf_xxx.ipynb

↥ back to top

TFLite Models

Click to expand

Under the scripts/inference/tflite/ folder, there are subfolders of different types of computer vision models application or configuration, in TFLite format:

• TFLite Model - Bounding Box

File Structure

Each of the subfolders contains a standard file structure. Below are descriptions for each file and folder regarding its content or purpose.

• input/: Some sample test images for prediction
• output/: Output folder to store predicted images
• requirements.txt: Python3 dependencies
• model_architecture/: All bounding box model architectures offered on Nexus platform
  • predict.py: Python script to load downloaded model and obtain predictions
  • tflite_xxx.ipynb: Jupyter notebook script to load model and obtain predictions

TFLite Model - Bounding Box

Click to expand

xxx in the subsequent command prompts represents chosen model architecture, which is also the name of the folders within the model_architecture/ folder.

Command to run Downloaded Model Python script

cd scripts/inference/tflite/bounding_box/

pip install -r requirements.txt

cd model_architecture/xxx

python predict.py --input "path_to_input_folder" --output "path_to_output_folder" --width 640 --height 640 --threshold 0.7 --model "path_to_model" --label "path_to_labelmap"

Sample default command

python predict.py --input "./input" --output "./output" --width 640 --height 640 --threshold 0.7 --model "./tf.lite" --label "./label_map.pbtxt"

Below is the list of modifiable script parameters and its description.

--input "path_to_input_folder" (Required)
--output "path_to_output_folder" (Required)
--model "path_to_model" (Required)
--label "path_to_labelmap" (Required)
--width "width of image to load" (Optional) (default: 640)
--height "height of image to load" (Optional) (default: 640)
--threshold "confidence threshold" (Optional) (default: 0.7)

Command to run Jupyter Notebook

pip install jupyter

python -m notebook tflite_xxx.ipynb

↥ back to top