Object detection using image classification models
In one aspect, the present disclosure relates to a method for or performing single-pass object detection and image classification. The method comprises receiving image data for an image in a system comprising a convolutional neural network (CNN), the CNN comprising a first convolutional layer, a last convolutional layer, and a fully connected layer; providing the image data to an input of the first convolutional layer; extracting multi-channel data from the output of the last convolutional layer; and summing the extracted data to generate a general activation map; and detecting a location of an object within the image by applying the general activation map to the image data.
1. A method for performing object detection using image classification models, the method comprising:
receiving image data for an image, wherein the image data is received in a system comprising a convolutional neural network (CNN), the CNN comprising an input layer, a first convolutional layer coupled to the input layer, a last convolutional layer, a fully connected layer coupled to the last convolution layer, and an output layer;
providing the image data to the input layer;
extracting multi-channel data from the last convolutional layer;
summing the multi-channel data to generate a two-dimensional general activation map; and
detecting a location of an object within the image by applying the general activation map to the image data.
2. The method of claim 1 wherein generating the general activation map comprises generating the general activation map without using class-specific weights.
3. The method of claim 1 wherein detecting the location of an object within the image comprises identifying a bounding box within the image based on comparing values within the general activation map to a predetermined threshold value.
4. The method of claim 1 wherein detecting the location of an object within the image comprises:
interpolating data within the general activation map; and
identifying a bounding box within the image using the interpolated data.
5. The method of claim 1 wherein detecting the location of an object within the image comprises upscaling the general activation map based on dimensions of the image.
6. A method for augmenting an image using single-pass object detection and image classification, the method comprising:
receiving image data for an image, wherein the image data is received in a system comprising a convolutional neural network (CNN), the CNN comprising an input layer, a first convolutional layer coupled to the input layer, a last convolutional layer, a fully connected layer coupled to the last convolution layer, and an output layer;
extracting multi-channel data from the output of the last convolutional layer;
summing the extracted data to generate a two-dimensional general activation map;
detecting a location of an object within the image by applying the general activation map to the image data;
receiving one or more classifications the output layer; and
displaying the image and a content overlay, wherein a position of the content overlay relative to the image is determined using the detected object location, wherein the content overlay comprises information determined by the one or more classifications.
7. The method of claim 6 wherein generating the general activation map comprises generating the general activation map without using class-specific weights.
8. The method of claim 6 wherein detecting the location of an object within the image comprises identifying a bounding box within the image based on comparing values within the general activation map to a predetermined threshold value.
9. The method of claim 6 wherein detecting the location of an object within the image comprises:
interpolating data within the general activation map; and
identifying a bounding box within the image using the interpolated data.
10. The method of claim 6 wherein detecting the location of an object within the image comprises upscaling the general activation map based on dimensions of the image.
11. A system for performing single-pass object detection and image classification, the system comprising:
a processor;
a convolutional neural network (CNN) configured for execution on the processor, the CNN comprising a first convolutional layer, a last convolutional layer, and a fully connected layer, wherein an output of the last convolutional layer is coupled to an input of the fully connected layer;
an image ingestion module configured for execution on the processor to receive image data for an image and to provide the image data to an input of the first convolutional layer;
an object detection module configured to extract multi-channel data from the output of the last convolutional layer, sum the extracted data to generate a two-dimensional general activation map, and to detect a location of an object within the image by applying the general activation map to the image data; and
an image augmentation module configured for execution on the processor to receive one or more classifications from an output of the fully connected layer and to display the image and a content overlay, wherein a position of the content overlay relative to the image is determined using the detected object location.
12. The system of claim 11 wherein the object detection module is configured to generate the general activation map without using class-specific weights.
13. The system of claim 11 wherein the object detection module is configured to detect the location of an object within the image by identifying a bounding box within the image based on comparing values within the general activation map to a predetermined threshold value.
14. The system of claim 11 wherein the object detection module is configured to:
interpolate data within the general activation map; and
identify a bounding box within the image using the interpolated data.
15. The system of claim 11 wherein the object detection module is configured to detect the location of an object within the image by upscaling the general activation map based on dimensions of the image.
16. A non-transitory computer-readable medium storing program instructions that are executable to:
receive image data for an image, wherein the image data is received in a system comprising a convolutional neural network (CNN), the CNN comprising a first convolutional layer, a last convolutional layer, and a fully connected layer, wherein an output of the last convolutional layer is coupled to an input of the fully connected layer;
provide the image data to an input of the first convolutional layer;
extract multi-channel data from the output of the last convolutional layer;
sum the extracted data to generate a two-dimensional general activation map;
detect a location of an object within the image by applying the general activation map to the image data;
receive one or more classifications from an output of the fully connected layer; and
display the image and a content overlay, wherein a position of the content overlay relative to the image is determined using the detected object location.
17. The non-transitory computer-readable medium of claim 16 wherein the program instructions are executable to generate the general activation map without using class-specific weights.
18. The non-transitory computer-readable medium of claim 16 wherein the program instructions are executable to detect the location of an object within the image by identifying a bounding box within the image based on comparing values within the general activation map to a predetermined threshold value.
19. The non-transitory computer-readable medium of claim 16 wherein the program instructions are executable to:
interpolate data within the general activation map; and
identify a bounding box within the image using the interpolated data.
20. The non-transitory computer-readable medium of claim 16 wherein the program instructions are executable to detect the location of an object within the image by upscaling the general activation map based on dimensions of the image.