Attribute-aware outfit recommendation

1. A method, comprising:

receiving, by a computing device, first image data depicting a first fashion item;

receiving, by the computing device, first category data representing a first item category, the first item category including a first plurality of fashion items;

receiving, by the computing device, a first attribute vector representing a selected fashion item attribute, the first attribute vector comprising a first number of dimensions representing color and a second number of dimensions representing fabric;

generating, by the computing device, a first output feature vector based on the first image data, the first category data, and the first attribute vector;

comparing the first output feature vector to a plurality of other feature vectors, each of the plurality of other feature vectors representing a respective fashion item of the first plurality of fashion items of the first item category;

selecting a feature vector representing a second fashion item based at least in part on a comparison of the first output feature vector and the feature vector representing the second fashion item; and

outputting second image data representing the second fashion item, wherein the second fashion item includes the selected fashion item attribute.

2. The method of claim 1 , further comprising:

determining respective attributes of each of the first plurality of fashion items of the first item category by applying named entity recognition to respective descriptions of each of the first plurality of fashion items; and

generating, for each of the first plurality of fashion items, respective attribute vectors representing the respective attributes, wherein the plurality of other feature vectors represent the respective attributes and respective appearances of each of the first plurality of fashion items.

3. The method of claim 1 , further comprising:

generating a state vector based on the receiving, by the computing device, the first image data depicting the first fashion item; and

outputting, by the computing device, the second image data based at least in part on the state vector.

4. The method of claim 1 , wherein generating the first output feature vector comprises using a long short-term memory (LSTM) network of the computing device to generate the first output feature vector, the method further comprising:

generating, by the LSTM network of the computing device, a state vector, wherein the state vector represents an outfit comprising at least the first fashion item and the second fashion item.

5. The method of claim 1 , wherein the first item category comprises a next item category requested for recommendation by the computing device, and the selected fashion item attribute represents a requested attribute for the second image data.

6. The method of claim 5 , further comprising:

combining a visual feature representation of the first image data, the first category data, and the first attribute vector to generate a combined input vector; and

inputting the combined input vector into an long short term memory (LSTM) cell.

7. The method of claim 1 , further comprising:

providing the first image data to a visual feature extractor function; and

generating, by the visual feature extractor function, an input feature vector representing the first image data, wherein the first output feature vector is generated further based on the input feature vector.

8. The method of claim 7 , further comprising:

combining the input feature vector, the first attribute vector, and a category vector representing the first item category to generate a combined input vector; and

providing the combined input vector to a machine learning model, wherein the first output feature vector is generated based on the combined input vector and a first state vector.

9. The method of claim 1 , wherein the first attribute vector represents at least a selected color and a selected fabric for a recommended fashion item.

10. A computing device, comprising:

at least one processor; and

non-transitory computer-readable memory configured to be in communication with the at least one processor, the non-transitory computer-readable memory effective to store instructions that, when executed by the at least one processor, cause the at least one processor to:

receive, by the computing device, first image data depicting a first fashion item;

receive, by the computing device, first category data representing a first item category, the first item category including a first plurality of fashion items;

receive, by the computing device, a first attribute vector representing a selected fashion item attribute, the first attribute vector comprising a first number of dimensions representing color and a second number of dimensions representing fabric;

generate, by the computing device, a first output feature vector based on the first image data, the first category data, and the first attribute vector;

compare the first output feature vector to a plurality of other feature vectors, each of the plurality of other feature vectors representing a respective fashion item of the first plurality of fashion items of the first item category;

select a feature vector representing a second fashion item based at least in part on a comparison of the first output feature vector and the feature vector representing the second fashion item; and

output second image data representing the second fashion item, wherein the second fashion item includes the selected fashion item attribute.

11. The computing device of claim 10 , the non-transitory computer-readable memory storing further instructions that, when executed by the at least one processor, are further effective to cause the at least one processor to:

determine respective attributes of each of the first plurality of fashion items of the first item category by applying named entity recognition to respective descriptions of each of the first plurality of fashion items; and

generate, for each of the first plurality of fashion items, respective attribute vectors representing the respective attributes, wherein the plurality of other feature vectors represent the respective attributes and respective appearances of each of the first plurality of fashion items.

12. The computing device of claim 10 , the non-transitory computer-readable memory storing further instructions that, when executed by the at least one processor, are further effective to cause the at least one processor to:

generate a state vector based on the receiving the first image data depicting the first fashion item; and

output the second image data based at least in part on the state vector.

13. The computing device of claim 10 , the non-transitory computer-readable memory storing further instructions that, when executed by the at least one processor, are further effective to cause the at least one processor to:

combine a visual feature representation of the first image data, the first category data, and the first attribute vector to generate a combined input vector; and

input the combined input vector into an long short term memory (LSTM) cell.

14. The computing device of claim 10 , wherein the first item category comprises a next item category requested for recommendation by the computing device, and the selected fashion item attribute represents a requested attribute for the second image data.

15. The computing device of claim 10 , the non-transitory computer-readable memory storing further instructions that, when executed by the at least one processor, are further effective to cause the at least one processor to:

provide the first image data to a visual feature extractor function; and

generate, by the visual feature extractor function, an input feature vector representing the first image data, wherein the first output feature vector is generated further based on the input feature vector.

16. The computing device of claim 15 , the non-transitory computer-readable memory storing further instructions that, when executed by the at least one processor, are further effective to cause the at least one processor to:

combine the input feature vector, the first attribute vector, and a category vector representing the first item category to generate a combined input vector; and

provide the combined input vector to a machine learning model, wherein the first output feature vector is generated based on the combined input vector and a first state vector.

17. The computing device of claim 10 , wherein the first attribute vector represents at least a selected color and a selected fabric for a recommended fashion item.

18. A method comprising:

receiving, by a computing device, first image data depicting a first fashion item;

receiving, by the computing device, first category data representing a first item category, the first item category including a first plurality of fashion items;

receiving, by the computing device, a first attribute vector representing a selected fashion item attribute, the first attribute vector comprising a first number of dimensions representing color and a second number of dimensions representing fabric;

generating, by the computing device, a first output feature vector based on the first image data, the first category data, and the first attribute vector;

comparing the first output feature vector to a plurality of other feature vectors, each of the plurality of other feature vectors representing a respective fashion item of the first plurality of fashion items of the first item category;

selecting, from among the plurality of other feature vectors, a feature vector representing a second fashion item; and

outputting second image data representing the second fashion item, wherein the second fashion item includes the selected fashion item attribute.

19. The method of claim 18 , further comprising:

determining respective attributes of each of the first plurality of fashion items of the first item category by applying named entity recognition to respective descriptions of each of the first plurality of fashion items; and

generating, for each of the first plurality of fashion items, respective attribute vectors representing the respective attributes, wherein the plurality of other feature vectors represent the respective attributes and respective appearances of each of the first plurality of fashion items.

20. The method of claim 18 , further comprising:

generating a state vector based on the receiving, by the computing device, the first image data depicting the first fashion item; and

outputting, by the computing device, the second image data based at least in part on the state vector.