This is the readme file to provide the links to other key data that is used in our ECCV 2014 embedding paper as well as our TPAMI paper (if accepted). 1, 100 dimensional skip-gram model: this can be download from http://www.eecs.qmul.ac.uk/~yf300/embedding/textw2v_wiki100_skip.bin (1.79G) The skip-gram model is trained by using the codes of google word2vec with all English wikipedia articles (explained in our papers). The idea is that with the trained model, you get one input word, and it will return a 100-dimensional vector. One important trick of normalisation is that you had to do l2-normalisation to each word vector. In my applications, that is important to get reasonable results! 2, I provide my code: http://www.eecs.qmul.ac.uk/~yf300/embedding/final.o (linux version) http://www.eecs.qmul.ac.uk/~yf300/embedding/word2vecQuery.rar to generate all the words vectors from the dictionary: The invoke format is : ./final.o textw2v_wiki100_skip.bin my-word-list.txt The my-word-list.txt should be like this: Roman Italy birthday_party hello_world giant panda And ./final.o will generate result.txt at the same folder and each row of result.txt will correspond to the 100 dimensional word vector for each corresponding row of my-word-list.txt 3, I will release the data used for our PAMI submission if it is accepted.