File Descriptions
images_train
- 10,000 training images of size 224x224.
images_test
- 2,000 test images of size 224x224.
tags_train
- image tags correspond to training images. Each image have several tags indicating the human-labeled object categories appear in the image, in the form of "supercategory:category".
tags_test
- image tags correspond to test images. Each image have several tags indicating the human-labeled object categories appear in the image, in the form of "supercategory:category".
features_train
- features extracted from a pre-trained Residual Network (ResNet) on training set, including 1,000 dimensional feature from classification layer (fc1000) and 2,048 dimensional feature from final convolution layer (pool5). Each dimension of the fc1000 feature corresponds to a WordNet synset
[here](https://gist.github.com/yrevar/942d3a0ac09ec9e5eb3a)
.
features_test
- features extracted from the same Residual Network (ResNet) on test set, including 1,000 dimensional feature from classification layer (fc1000) and 2,048 dimensional feature from final convolution layer (pool5).
descriptions_train
- image descriptions correspond to training images. Each image have 5 sentences for describing the image content.
descriptions_test
- image descriptions for test images. Each image have 5 sentences for describing the image content.
Notice that one test description corresponds to one test image. The task you need to do is to return top 20 images in test set for each test description.
sample_submission.csv
- a sample submission file in the correct format.