ANN Training for Custom Object Detection
- Product Code: CT_04
- Availability: In Stock
Custom training is ideal for businesses with the need to quickly classify image or video content.
It is using TIGER ENGINEERING's auto categorization technology for very specific set of image categories. For this we need the structure of your categories and a list of public image URLs, or alternatively an archive of images, that belong to each category. Ideally we need 500-1000+ representative images for each category. It’s essential that you design the structure of your categories in way that they are distinct and non-overlapping.
Depending on the number of the categories and the sample photos the whole process may take up to 10 business days to finish.
|We aim to achieve the highest possible precision rates. We very often get to the 85-95% range in terms of precision rates, depending on the complexity of the task. After reviewing the training data set, we typically talk with the client and agree on an acceptable precision rate that would ensure the business goals of the clients are met.||We charge 100 € upfront in order to review the input data and run the custom training process. Once the training is completed we share the validation statistics. If the results met your needs and you want to go in production we charge a pre-agreed success fee starting from 600 €. The success fee varies depending on the complexity of the training task. You will receive a quote before we start working on your case.||We work to provide state-of-the-art image recognition technologies.One of the hardest task in the image recognition technology is to have enough good data so that we can train the network. As a reward for helping us learn from your data we can offer discount options starting from 20% of the training price. If you allow us, we will store the image samples and continue to use them solely for the purpose of improving our technology. This means that in any case we WON’T provide any of the image data to third-parties, competitors or advertisers. The percentage of the discount we can offer depends on the quality of the data you have.|
The number of photos in each category don't need to be exactly the same, but ideally there won’t be more than x2 times difference between the smallest and largest number of sample photos for their respective categories.What if I am not satisfied with the precision of the results?
There are several options - adding more diverse sample photos per category can significantly increase the precision rate. Sometimes categories that are overlapping could be the reason for less than optimal results. In very rare cases we may jointly conclude that we can’t do anything feasible at this stage and you don’t need to pay the success fee. We try to minimize this case by carefully analyzing the definition of the categorization task before we start the actual training and request the upfront fee.