Dbcollection

This release brings some breaking changes compared to the previous release (`0.1.7`), but those were necessary and important, bringing extra functionality as well as some bug fixes. Now, users should have a more powerful API to help them deal with data loading and setup.

New features
- add method to reload the cache contents (41)
- new functionality added to the data loader object (40)

Changes
- refactor url download(35)
- remove osx from travis for faster tests (36)
- refactor how datasets are setup (38)
- Reorganize where third-party modules are stored (39, 46)
- Refactor how data is loaded and how samples are fetched (40)
- Refactor how metadata is saved to HDF5 files (40)
- Change `cache_dir` with `download_dir` path for storing downloaded files (44)
- Update docs (46)
- Move dataset managing API methods into a single file (50)

Bug fixes
- Fix ucf sports video extraction (33)
- Fix some datasets urls (44)
- Fix imports (49)
- Other small bug fixes

New features
- Added `.info()` method to the data loader API. https://github.com/dbcollection/dbcollection/commit/3ea6aa2744435800cf29d7f4ca0a2ec793949475

Changes
- Refactor the `info()` method to display more information about the used and/or available datasets in cache.
- Updated documentation. https://github.com/dbcollection/dbcollection/commit/ab3907eaa31b345a8bc498f714ec1ece31fffe38
- (db) Set the random seed for the ucf sports recognition (https://github.com/dbcollection/dbcollection/commit/c3e3c02f7bbcf7b1943a50be880100d4a037902c) and detection (https://github.com/dbcollection/dbcollection/commit/978ec2ec4890aef0ada7510cc3badc0ef404b8d0) tasks.
- (db) Modified ucf sports dataset. https://github.com/dbcollection/dbcollection/commit/be5792b86af71e043e6bd1e60b2abea86c131bbd
- The default task is now translated to the correct associated task. https://github.com/dbcollection/dbcollection/commit/6bd32a8dc4bd1f84eb36b394c0339facf5fcd870
- Remove unnecessary tasks from the coco dataset. https://github.com/dbcollection/dbcollection/commit/01c2d18038dd7bf30c20b0125a974006e2ca2336

Bug fixes
- Fixed docstrings. https://github.com/dbcollection/dbcollection/commit/f636f0ca9358682a5c6f80a3f88e304ecf8374c9
- Object method should accept empty indexes. https://github.com/dbcollection/dbcollection/commit/050c21f0c70e5c328aec2d37a7696ed0aedbcb3f
- (db) Caltech pedestrian detection task. https://github.com/dbcollection/dbcollection/commit/b419e3837409c2cac8c7521beb34bad5c21a6297

New features

- Tasks now default to discard the source group in the hdf5 files.
- Enable the source group (if exists) by adding the _s suffix to the task identifier. Metadata files should now occupy less space.
- Added unit tests to many core methods.

Changes

- Add extra functinality to the .info() method. Now it can select what information to display via input arguments.
- Refactored url downloading. Files now download to disk using a different name than the final one. Users that interrupts the downloading process now don't have to manually delete the empty/corrupted files left due to the interruption.
- Created the dbcollection organization and transfered all the code.
- Moved lua/torch api from the main repo to a separate one.
- Moved matlab api from the main repo to a separate one.
- Removed the pad_list2 method.
- Started adding some doctests to document some methods

Bug fixes
- Fix .object() method in DatasetLoader when inserting a list of indexes.
- Fix COCO dataset not loading properly with the code refactoring.
- Fix Lua/Torch and Matlab APIs to mirror the new API functionality.
- Lua/Torch API now allows the insertion of sequential variables instead of passing a table as argument.
- Fix load_pickle() method for Python 2.7

This release fixes some bugs related to 0.1.4 regarding python 2.7 compatability.

New features

- Matlab support
- Continuous integration support with travis, appveyor and codecov
- Initial documentation with readthedocs
- Added a class for better code reuse when testing datasets.
- Added conda support

Improvements

- some API changes for easier usage of the package: access the load/download/process/etc. methods is now set to the main module for easier use.

Bug fixes

- mostly fixes for some datasets processing