Until recently, IBM Watson Visual Recognition was mainly used to recognize an image as a whole. However, working with the picture as a whole is far from the most correct approach. Now, with a new feature object recognition, IBM Watson users got the opportunity to train models on images with marked objects, for their subsequent recognition on any frames.
Let's show how it can be done now.
If earlier with the help of IBM Watson you could distinguish a damaged car from an undamaged one, now you can not only recognize the presence of damage, but also evaluate its position and size. This approach is much more informative, allowing you to make predictions about the cost of the necessary repairs.
Of course, the list of options for using this functionality is much wider than a simple check of the integrity of the car. Now you can use Watson Visual Recognition to:
- Counting the number of people in queues or cars in traffic
- Identification of goods on the shelves
- Recognition of logos in photographs
- Analysis of CT and MRI images for abnormalities
- Other tasks related to working with specific objects in images
You don't have to spend months selecting and labeling data - our model has already been trained on several million samples and provides a fairly high quality prediction without any changes. If necessary, you can always retrain it so that the neural network meets the specifics of your field of activity.
Label images and train models on your data faster with Watson Studio
Usually, training your own model to accurately recognize objects is the most difficult task when creating a computer vision system. Watson Studio speeds up this process and helps you save time when working with large amounts of data. In conjunction with the free add-on Auto Label you can quickly mark up all the images in the dataset.
Beginning of work
After activating and creating a Visual Recognition application in the cloud, connect it to Watson Studio and in the Custom Models section create a model in the Detect Objects window.
Upload your raw data to Watson Studio (JPEG, PNG or ZIP archive containing these images can be used)
Select an image, select the object you want to recognize, give it a name and save. Repeat until you have selected all the necessary objects in this image.
Once you mark up a few images, you can train and test your model.
You can also add more images to improve the quality of your model using the Auto Label feature to help you label all your data. To use this feature, select all the necessary images and click on the "Auto Label" button so that Watson will label the data according to the specified classes.
After checking the accuracy of your model, you can embed the finished solution into your product.
object recognition with IBM Watson Visual Recognition for free today!
We would also like to invite you to free training seminars on и held in November in the client center of our Moscow office.
Additional materials:
- with cookies for training
Source: habr.com