Yes, our platform support importation of CSV file and json file format from Labelme.

Yes, you can download your annotations directly from the platform in CSV format.

Yes, you can , when creating a project, you can select any or all 3 type of annotation, and annonotate all at once as you go through each image.

All work will be automatically saved as soon as you annotate it, so you won't have to worry about loosing your valuable annotations. However, you must 'submit' your annotation in order fo it to be sent to training.

yes, you can configure the hotkeys based on your habit.

yes, touchscreen and stylus are supported.

The data used for training, should be as close to real world data as possible, so that the trained model will perform well in real world situation.

Our platform utilizes transfer learning and data augmentation so that we can reduce the amount of data needed; Even though having more data could lead to better deep learning results, but the quality of the data is just as important, if not more important.

We currently do now support uploading of models, if there such a need, please contact us.

We provide semi-manual hyper-parameter tuning for advanced users, who are familiar with the technical terms , so we do not provide addition instructions on this topic. A quick online search should provide you with what you need to learn about hyper-parameter turning.

Yes, after the training is completed, you can download the model, including all the parameters and quantitative analysis of the batch inference.

The downloaded model will include themodel architecture, model weight, inference script template, users will need expertise in deploying and configuring relevant parameters. DeepQ also provide model deployment services, please contact us for more details.

With image classificiation, our platform provide heat-map so that you can have a visual explanation of the inference result.

With more data available for training, the less likelihood of overfitting. Our platform utilizes data augmentation to help increase data number and have an early-stop mechanism to avoid over-fitting. We do recommend that you prepare as much training data as possible and the training data should be similar to the testing data.

Our platform is web-based, any computer or tablet with web-browser and a stable broadband internet connection can access our platform.

Our on-premise enterprise solution will include an all-in-one workstation with two Nvidia A100 GPU; however, if you already have GPU resources or require other specification or configuration, please contact us for more details.

Our cloud solutions meet all the information security requirements and DeepQ will at no instance or in any way access user data.

Our AI platform is currently focused in computer vision and we will continue to make platform enhancement centered around images. However, if there's an non-imaging project you are interested, please feel free to contact us, our team specializes in all aspects of artificial intelligence.

Our platform currently support png, jpg and DICOM images.