Virtual shoe try-on
A method includes extracting, using a backbone of a machine learning model, a plurality of features from an image of a foot and predicting, using a first portion of the machine learning model and based on one or more features of the plurality of features, a first aspect of the foot. The method also includes predicting, using a second portion of the machine learning model and based on one or more features of the plurality of features, a second aspect of the foot different from the first aspect, generating, using at least the first aspect and the second aspect, a two-dimensional model of a shoe, and superimposing the two-dimensional model of the shoe onto the image of the foot.
1. A method for fitting a virtual shoe onto a foot, the method comprising:
receiving, from a database, a three-dimensional model of a shoe;
receiving, at a backbone of a neural network, an image of a foot, wherein the image further includes at least one of a portion of a leg or a portion of clothing;
extracting, using the backbone of the neural network, a plurality of features from the image of the foot;
predicting, using a first subnetwork of the neural network and based on one or more features of the plurality of features, a first aspect of the foot;
predicting, using a second subnetwork of the neural network and based on one or more features of the plurality of features, a second aspect of the foot different from the first aspect, wherein the second aspect is a portion of the foot that will be occluded by at least one of the portion of the leg or the portion of clothing;
generating, using at least the first aspect, the second aspect, and the three-dimensional model of the shoe, a two-dimensional model of the shoe, wherein generating the two-dimensional model of the shoe comprises removing, from the three-dimensional model of the shoe, a portion of the three-dimensional model of the shoe corresponding to at least one of the portion of the leg or the portion of clothing; and
superimposing the two-dimensional model of the shoe onto the image of the foot.
2. The method of claim 1 , wherein the first aspect is an orientation of the foot in the image, and wherein generating the two-dimensional model of the shoe comprises rotating the three-dimensional model of the shoe or the two-dimensional model of the shoe to align the three-dimensional model of the shoe or the two-dimensional model of the shoe with the orientation of the foot in the image.
3. The method of claim 1 , further comprising predicting, using a third subnetwork of the neural network and based on one or more features of the plurality of features, whether the foot in the image is a right foot or a left foot.
4. A method comprising:
extracting, using a backbone of a machine learning model, a plurality of features from an image of a foot, wherein the image further includes at least one of a portion of a leg or a portion of clothing;
predicting, using a first portion of the machine learning model and based on one or more features of the plurality of features, a first aspect of the foot;
predicting, using a second portion of the machine learning model and based on one or more features of the plurality of features, a second aspect of the foot different from the first aspect, wherein the second aspect is a portion of the foot that will be occluded by at least one of the portion of the leg or the portion of the clothing;
generating, using at least the first aspect and the second aspect, a two-dimensional model of a shoe, wherein the generating the two-dimensional model of the shoe comprises removing, from a three-dimensional model of the shoe, a portion of the three-dimensional model of the shoe corresponding to the portion of the foot that will be occluded by at least one of the portion of the leg or the portion of the clothing; and
superimposing the two-dimensional model of the shoe onto the image of the foot.
5. The method of claim 4 , wherein the first aspect is an orientation of the foot in the image, and wherein generating the two-dimensional model of the shoe comprises rotating a three-dimensional model of the shoe or the two-dimensional model of the shoe according to the orientation.
6. The method of claim 4 , wherein the first portion comprises:
a feature extractor that generates a vector based on one or more features of the plurality of features;
a predictor that generates a heatmap based on the vector; and
a post processor that generates a prediction based on the heatmap.
7. The method of claim 4 , further comprising:
receiving a selection of the shoe; and
retrieving, from a database and based on the selection, a three-dimensional model of the shoe, wherein the two-dimensional model of the shoe is generated based on the three-dimensional model of the shoe.
8. The method of claim 4 , wherein the second aspect is whether the foot in the image is a right foot or a left foot.
9. The method of claim 4 , further comprising:
detecting a position of the foot in a captured image; and
cropping the captured imaged based on the detected position of the foot to produce the image of the foot.
10. The method of claim 4 , wherein generating the two-dimensional model of the shoe comprises re-sizing a three-dimensional model of the shoe according to one or more features of the plurality of features.
11. A system comprising:
a memory; and
a processor communicatively coupled to the memory, the processor configured to:
extract, using a backbone of a machine learning model, a plurality of features from an image of a foot, wherein the image further includes at least one of a portion of a leg or a portion of clothing;
predict, using a first portion of the machine learning model and based on one or more features of the plurality of features, a first aspect of the foot;
predict, using a second portion of the machine learning model and based on one or more features of the plurality of features, a second aspect of the foot different from the first aspect, wherein the second aspect is a portion of the foot that will be occluded by at least one of the portion of the leg or the portion of the clothing;
generate, using at least the first aspect and the second aspect, a two-dimensional model of a shoe, wherein the generating of the two-dimensional model of the shoe comprises removing, from a three-dimensional model of the shoe, a portion of the three-dimensional model of the shoe corresponding to the portion of the foot that will be occluded by at least one of the portion of the leg or the portion of the clothing; and
superimpose the two-dimensional model of the shoe onto the image of the foot.
12. The system of claim 11 , wherein the first aspect is an orientation of the foot in the image, and wherein generating the two-dimensional model of the shoe comprises rotating a three-dimensional model of the shoe or the two-dimensional model of the shoe according to the orientation.
13. The system of claim 11 , wherein the first portion comprises:
a feature extractor that generates a vector based on one or more features of the plurality of features;
a predictor that generates a heatmap based on the vector; and
a post processor that generates a prediction based on the heatmap.
14. The system of claim 11 , wherein the processor is further configured to:
receive a selection of the shoe; and
retrieve, from a database and based on the selection, a three-dimensional model of the shoe, wherein the two-dimensional model of the shoe is generated based on the three-dimensional model of the shoe.
15. The system of claim 11 , wherein the second aspect is whether the foot in the image is a right foot or a left foot.
16. The system of claim 11 , wherein the processor is further configured to:
detect a position of the foot in a captured image; and
crop the captured imaged based on the detected position of the foot to produce the image of the foot.
17. The system of claim 11 , wherein generating the two-dimensional model of the shoe comprises re-sizing a three-dimensional model of the shoe according to one or more features of the plurality of features.