Simultaneous object localization and attribute classification using multitask deep neural networks
Various embodiments are disclosed for simultaneous object localization and attribute classification using multitask deep neural networks. In an embodiment, a method comprises: obtaining, by a processing circuit, an image from an image capture device in an environment, the image including a target object in the environment; generating, by the processing circuit, predictions from the image for the target object using a multitask deep neural network, the multitask deep neural network including a network trunk and side branches, the network trunk configured for multi-scale feature extraction guided by supervision information provided by the side branches during training of the multitask deep neural network, the side branches configured as learning task-specific classifiers; and using, by the processing circuit, the predictions to localize the target object in the environment and to classify the target object and at least one attribute of the target object.
1. A method comprising:
obtaining, by a processing circuit, an image from an image capture device in a shopping environment, the image including a target object and at least one other object in the shopping environment;
generating, by the processing circuit, predictions from the image for the target object using a multitask deep neural network, the multitask deep neural network including a network trunk and side branches, the network trunk configured for multi-scale feature extraction guided by supervision information provided by the side branches during training of the multitask deep neural network, the side branches including at least a target object classification branch configured to predict a class of the target object, a target object localization branch configured to predict a location of the target object in the shopping environment and a target attribute prediction branch configured to predict at least one target attribute associated with a pattern of behavior of the target object with respect to the at least one other object at the predicted location; and
using, by the processing circuit, the predictions to localize the target object in the environment, identify the target object, and detect the pattern of behavior of the identified target object relative to the at least one other object at the location in the shopping environment.
2. The method of claim 1 , wherein the image is taken from a camera in a retail environment, the target object is classified as a shopper and the pattern of behavior of the shopper in the shopping environment is theft of the at least one other object.
3. The method of claim 1 , wherein the side branches are connected to multi-scale feature maps for detecting target objects of different sizes.
4. The method of claim 1 , where the multitask deep neural network predicts multifaceted attribute labels for the target object.
5. The method of claim 1 , wherein the multitask deep neural network is a single feedforward neural network.
6. The method of claim 1 , wherein the multitask deep neural network implements a multitask objective function for joint many-task optimization.
7. The method of claim 6 , wherein the multitask objective function is a loss function that includes a weighted sum of localization loss, classification loss and attribute prediction loss.
8. The method of claim 1 , wherein the bounding box regression branch regresses shape offsets of a set of predefined default anchor boxes with respect to a ground truth box, and default anchor boxes of different sizes and aspect ratios are tiled on each cell of a feature map provided by the network trunk, and wherein the position of each anchor box relative to its corresponding cell is fixed.
9. The method of claim 8 , wherein the classification branch produces confidence scores of target object classes for each of the default anchor boxes.
10. A system comprising:
one or more processors;
memory storing instructions that when executed by the one or more processors, cause the one or more processors to perform operations comprising:
obtaining an image from an image capture device in a shopping environment, the image including a target object and at least one other object in the shopping environment;
generating predictions from the image for the target object using a multitask deep neural network, the multitask deep neural network including a network trunk and side branches, the network trunk configured for multi-scale feature extraction guided by supervision information provided by the side branches during training of the multitask deep neural network, the side branches including at least a target object classification branch configured to predict a class of the target object, a target object localization branch configured to predict a location of the target object in the shopping environment and a target attribute prediction branch configured to predict at least one target attribute associated with a pattern of behavior of the target object relative to the at least one other object at the predicted; and
using the predictions to localize the target object in the shopping environment, identify the target object, and detect the pattern of behavior behaviour of the target object relative to the at least one other object at the location in the shopping environment.
11. The system of claim 10 , wherein the image is taken from a camera in shopping environment, the target object is classified as a shopper and the pattern of behavior of the shopper in the shopping environment is theft of the at least one other object.
12. The system of claim 10 , wherein the side branches are connected to multi-scale feature maps for detecting target objects of different sizes.
13. The system of claim 10 , where the multitask deep neural network predicts multifaceted attribute labels for the target object.
14. The system of claim 10 , wherein the multitask deep neural network is a single feedforward neural network.
15. The system of claim 10 , wherein the multitask deep neural network implements a multitask objective function for joint many-task optimization.
16. The system of claim 15 , wherein the multitask objective function is a loss function that includes a weighted sum of localization loss, classification loss and attribute prediction loss.
17. The system of claim 10 , wherein the bounding box regression branch regresses shape offsets of a set of predefined default anchor boxes with respect to a ground truth box, and default anchor boxes of different sizes and aspect ratios are tiled on each cell of a feature map provided by the network trunk, and wherein the position of each anchor box relative to its corresponding cell is fixed.
18. The system of claim 17 , wherein the classification branch produce confidence scores of target object classes for each of the default anchor boxes.
19. A system comprising:
an image capture device;
a database;
a retail analytics processing circuit coupled to the image capture device, the processing circuit configured to:
obtain, from the image capture device, an image of a shopping environment, the image including a target object and at least one other object;
generate predictions from the image for the target object using a multitask deep neural network, the multitask deep neural network including a network trunk and side branches, the network trunk configured for multi-scale feature extraction guided by supervision information provided by the side branches during training of the multitask deep neural network, the side branches including at least a target object classification branch configured to predict a class of the target object, a target object localization branch configured to predict a location of the target object in the shopping environment and a target attribute prediction branch configured to predict at least one target attribute associated with a behavior of the target object relative to the at least one other object at the predicted location;
use the predictions to determine the location of the target object in the shopping environment, identify the target object as a customer or employee, and detect the pattern of behavior of the identified customer or employee relative to the at least one other object at the predicted location in the shopping environment; and
update one or more records in the database that are related to the customer or employee.
20. The method of claim 1 , wherein the image is taken from a camera in a retail environment and the target object is classified based on demographics.
21. The system of claim 10 , wherein the image is taken from a camera in a retail environment and the target object is classified based on demographics.
22. The system of claim 19 , wherein the image is taken from a camera in a retail environment and the target object is classified based on demographics.
23. The method of claim 1 , wherein the image is taken from a camera in a retail environment and the target object is classified as standing, sitting or reaching.
24. The system of claim 10 , wherein the image is taken from a camera in a retail environment and the target object is classified as standing, sitting or reaching.
25. The system of claim 19 , wherein the image is taken from a camera in a retail environment and the target object is classified as standing, sitting or reaching.