Xtreme1 v0.6.0 : Support integrating user’s models and data import & export function enhancement

We are pleased to announce the release of Xtreme1 v0.6. After a couple of weeks of dedicated work, this update is now available for download.

In this update, Xtreme1 provides users with the ability to integrate their own model. Additionally, the display effect of the similarity map has been improved.

We understand that some users face challenges with data formats when uploading and exporting data. To solve this problem, we have updated our documentation, which can be found in Docs and the "universal" converter from Xtreme1-SDK.

New Features

Model Integration

To give users more control over their own models, we have added functionality that allows them to create, edit, search, and delete models. With this feature, users can manage basic model information, classes, configuration details, and run the model on datasets.

For more information, please refer to our Model Integration Guide.

Enhanced Function of Data Import and Export

Data Import

• Support for importing pre-annotation dataset on 3D Dataset

Note: Only importing the Class name is supported, Attributes and global Classification are not yet supported.

Note: Don’t forget to check “Contain annotation result”.

Data Export

• COCO format can be exported directly
• When exporting data, the Filter can be applied.

An example of COCO format JSON file:

{
 "info": {
  "contributor": "",
  "date_created": "2023-04-11T09:03:50Z",
  "description": "Basic AI Xtreme1 dataset 0.6 demo exported to COCO format (https://github.com/basicai/xtreme1)",
  "url": "https://github.com/basicai/xtreme1",
  "year": "2023",
  "version": "Xtreme1 V0.6"
 },
 "licenses": [],
 "images": [
  {
   "id": 0,
   "license": 0,
   "file_name": "filename.png",
   "xtreme1_url": "https://.../image_url",
   "width": 1024,
   "height": 1024,
   "date_captured": null
  }
 ],
 "annotations": [
  {
   "id": 0,
   "image_id": 0,
   "category_id": 1,
   "segmentation": [],
   "area": 679272,
   "bbox": [
    64,
    76,
    913,
    744
   ],
   "iscrowd": 0,
   "score": null
  }
 ],
 "categories": [
  {
   "id": 1,
   "name": "sea_otter",
   "supercategory": "",
   "attributes": {}
  }
 ]
}

Image Similarity Map Optimization

• The algorithm is optimized.
• Selecting multiple targets in the image similarity map, you can jump to the data tab to find the corresponding images.

Add Dataset Split

In the data preparation process of model training, the data can be directly divided into sub-datasets of Training, Test, and Validation sets.

A training dataset is a set of examples used to fit the parameters of a machine learning model, typically involving supervised learning methods and parameter adjustments. A validation dataset provides an unbiased evaluation of a model fit on the training dataset while tuning the model’s hyper-parameters, often used for regularization by early stopping. The test dataset offers an unbiased evaluation of a final model fit on the training dataset, ensuring that the model performs well on unseen data.

Changes and Bug Fixes

• Optimized the clarity of the preview image, including in dataset and scenario search
• In Scenario Search, the search box style is changed.
• Fixed the result could not be calculated in some cases.

Visit Xtreme1 Docs to learn more, or head over to the GitHub repo to install the updated Xtreme1. As always, if you have any feedback or suggestions for improvement, feel free to discuss them in our Slack community or let us know by opening an issue on GitHub, and we’ll be sure to help you out.

GitHub: https://github.com/xtreme1-io/xtreme1

Docs: https://docs.xtreme1.io

Slack: https://xtreme1io.slack.com

Website: https://www.xtreme1.io

Developed by BasicAI Team