“Image Retrival With Text Query” https://github.com/frankhlchi/Image-Search-Engine/blob/master/report.pdf
Files needed to run the code:
-
GoogleNews-vectors-negative300.bin.gz
-
unzip cs5785-fall-2019-final.zip (download from Kaggle, https://www.kaggle.com/c/cs5785-fall-2019-final/data)
-
(1)run corresponding resnet feature generation code (Team Ground Truth 1.extract resnet features part.ipynb); (2)OR you can also download ready-to-use self-generated features data directly (generated_resnet_features.zip), from https://drive.google.com/drive/folders/1pkXLFcvuEkC_VQ-QunVUC9EjjlvF8-j5, then unzip generated_resnet_features.zip in this folder
The File tree before running the model notebook
-
GoogleNews-vectors-negative300.bin.gz -/descriptions_train -/descriptions_test -/tags_train -/tags_test -/features_train -/features_test -/images_train -/images_test -resnet50_train.csv -resnet101_train.csv -resnet152_train.csv -resnet_ResNext_train.csv -resnet_wide101_train.csv -resnet50_test.csv -resnet101_test.csv -resnet152_test.csv -resnet_ResNext_test.csv -resnet_wide101_test.csv
-
GBM models are NOT required to run the best model. It is only used in other models. If you want to run them, you can (1)run corresponding lightGBM generation code; (2) download from https://drive.google.com/drive/folders/1pkXLFcvuEkC_VQ-QunVUC9EjjlvF8-j5 (unzip GBM models.zip into fold GBM_model)
-
If there is any further question please contact Hongliang CHI, hc962@cornell.edu; Kai Zhang kz298@cornell.edu