General Questions:
\r\nSo since their are 2 types either "devel" or "valid" and 20 choies for each of those, there would be a grand total of 40 batches with each batch having 100 gestures?
\r\nSo by saying that their are "There are instances of N unique gestures from a vocabulary of 8 to 15 gestures" you are saying that each batch contains a unique subset combination of the 8-15 gestures? Actually I think it would be permutation since order does\r\n matter(what order the gestures are done in)?
\r\nJust wanted to confirm that I do not have to publish any papers on this topic correct seeing as to how the deadlines for submission are before final evaluations?
\r\nTraining Data/Test Data
\r\nAm I allowed to use cross validation or do I have to keep data as training data and testing data seperate?
\r\nAlong the same lines am I allowed to use methods such as boosting/bagging?
\r\nI don't really understand how to read the training data/test data:
\r\nEx: In "devel01_train.csv", cell A1 specifies following:
\r\n| \r\n devel01_1,10 \r\n | \r\n
Should I go ahead and split the data on the commas, because in excel all the data is shown in one column?
\r\nSo devel01_1 would be the row id which wold be unique and 10 would be a label? What is this label? Is that label the classification?
\r\nThanks!
", "messages": [ { "id": "7228", "postDate": "12/16/2011 06:54:13", "content": "General Questions:
\r\nSo since their are 2 types either "devel" or "valid" and 20 choies for each of those, there would be a grand total of 40 batches with each batch having 100 gestures?
\r\nSo by saying that their are "There are instances of N unique gestures from a vocabulary of 8 to 15 gestures" you are saying that each batch contains a unique subset combination of the 8-15 gestures? Actually I think it would be permutation since order does\r\n matter(what order the gestures are done in)?
\r\nJust wanted to confirm that I do not have to publish any papers on this topic correct seeing as to how the deadlines for submission are before final evaluations?
\r\nTraining Data/Test Data
\r\nAm I allowed to use cross validation or do I have to keep data as training data and testing data seperate?
\r\nAlong the same lines am I allowed to use methods such as boosting/bagging?
\r\nI don't really understand how to read the training data/test data:
\r\nEx: In "devel01_train.csv", cell A1 specifies following:
\r\n| \r\n devel01_1,10 \r\n | \r\n
Should I go ahead and split the data on the commas, because in excel all the data is shown in one column?
\r\nSo devel01_1 would be the row id which wold be unique and 10 would be a label? What is this label? Is that label the classification?
\r\nThanks!
", "rawMarkdown": "", "votes": null }, { "id": "7248", "postDate": "12/16/2011 19:16:00", "content": "Data:
\r\nYou are now seeing a subset of the data, we will be releasing shortly more develxx batches for development. For final testing we will release finalxx batches.
\r\nAll the batches are organized in the same way. The all contain:
\r\n- 47 M files (RGB videos) and 47 corresponding K files (depth videos)
\r\n- The video contain sequences of 1 to 5 recorded gestures. In total 100 gestures are recorded.
\r\n- The valid and final batches contain labels for the first N videos used as training examples. The devel batches also contain the labels of the other videos (the test examples).
\r\nThe number N maybe different for each batch. It is between 8 and 15. It corresponds to the number of gesture tokens (unique gestures) in the lexicon associated to that batch (different gestures are played in different batches, e.g., one has diving signals,\r\n another has referee signals, etc.). The gestures are played in the various videos following a script that prescribed to the users in which order to play them. The script was obtained by drawing at random labels from the N possible values.
\r\nPublication:
\r\nPublishing methods is optional, we just provide this opportunity ro the participants.
\r\nDevelopment data:
\r\nYou are allowed to use the develxx files in any way you want. In particular, you can do cross-validation. We provide a split between training and test data to make the devel batches look like the valid and final batches. But this split does not need to be\r\n kept for your purposes. I am not sure how you want to use boosting/bagging, but feel free to try. In the end, the task will be:
\r\nGiven new batches from new lexicons of gestures you have not seen before, train on the N first examples and make predictions on the remainder.
\r\nCSV file format:
\r\n- First column => the ID of the example. devel01_1 means the 1st example of the batch devel01. valid04_12 means the 12th example of the batch valid04.
\r\n- Second column => the list of gestures that were played (space separated)
\r\n", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 7248, "author_name": "challengeadmin", "author_url": "", "post_date": "12/16/2011 19:16:00", "content": "
Data:
\r\nYou are now seeing a subset of the data, we will be releasing shortly more develxx batches for development. For final testing we will release finalxx batches.
\r\nAll the batches are organized in the same way. The all contain:
\r\n- 47 M files (RGB videos) and 47 corresponding K files (depth videos)
\r\n- The video contain sequences of 1 to 5 recorded gestures. In total 100 gestures are recorded.
\r\n- The valid and final batches contain labels for the first N videos used as training examples. The devel batches also contain the labels of the other videos (the test examples).
\r\nThe number N maybe different for each batch. It is between 8 and 15. It corresponds to the number of gesture tokens (unique gestures) in the lexicon associated to that batch (different gestures are played in different batches, e.g., one has diving signals,\r\n another has referee signals, etc.). The gestures are played in the various videos following a script that prescribed to the users in which order to play them. The script was obtained by drawing at random labels from the N possible values.
\r\nPublication:
\r\nPublishing methods is optional, we just provide this opportunity ro the participants.
\r\nDevelopment data:
\r\nYou are allowed to use the develxx files in any way you want. In particular, you can do cross-validation. We provide a split between training and test data to make the devel batches look like the valid and final batches. But this split does not need to be\r\n kept for your purposes. I am not sure how you want to use boosting/bagging, but feel free to try. In the end, the task will be:
\r\nGiven new batches from new lexicons of gestures you have not seen before, train on the N first examples and make predictions on the remainder.
\r\nCSV file format:
\r\n- First column => the ID of the example. devel01_1 means the 1st example of the batch devel01. valid04_12 means the 12th example of the batch valid04.
\r\n- Second column => the list of gestures that were played (space separated)
\r\n", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "7228": "", "7248": "" }, "source": "meta" }