Skip to content

Object detection with pre-trained Faster-RCNN

In this example we will consider applying trained Faster-RCNN model to detect objects.

To run inference you'll need to have some connected node with agent to use it computational resources.

Step 1: Get model

Add pre-trained model from Model Zoo if you haven't one yet. Open "Model Zoo", find "Faster-RCNN" preset, click "Add model" button and it would appear in My Models page.

The model is pretrained on MS COCO dataset.

Step 2. Select model

Find model in My Models table and click "Test" button.

Step 3: Select project

Select dataset or whole project you have uploaded before.

TODO: show 'demo' dataset?

Step 4: Determine settings

TODO: select 1) 2) 3) on image

4.1: Make up output project name

New project with inference results will be created after applying model.

4.2: Select node to run inference on it

Add node if you haven't one yet.

4.3: Tune settings for your inference

Fast inference

There are reasonable default values for every option, so one may start inference with empty {} settings.

Select provided config (named like faster_rcnn_full_image_all_classes) from drop-down list with saved inference configs.

Here are the simple settings:

{
  "gpu_devices": [
    0
  ],
  "model_classes": {
    "save_classes": "__all__",
    "add_suffix": "_bbox"
  },
  "existing_objects": {
    "save_classes": [],
    "add_suffix": ""
  },
  "mode": {
    "source": "full_image"
  }
}

Set device index in gpu_devices if you have multi-GPU node and some GPUs are busy. Please note that right now we support inference with only single GPU.

See corresponding section for description of the settings above and different inference modes.

In field "Model config" below you can see list of classes on which the model has been trained.

Step 5: Run task

Now click "Start inference" button.

Task list will be opened with the new inference task.

When task will be Done link to result project will appear.

Enjoy results.