fuel
fuel copied to clipboard
mila-iqia
A data pipeline framework for machine learning
Hello, I currently deal with image datasets of about 1 million images. When saving them as bumpy array (with dtype uint8) this would result in a dataset file of over...
Hi, thanks for this great work. I want to run the block example that do word reversion. I have read the documentation and download the [corpus tar](http://statmt.org/wmt11/training-monolingual.tgz). I untar it...
I propose to add a syntactic sugar to easier manipulate streams, we could use some infix operator to combine transformers (like `|` or `>`). `|` would look like linux piping:...
When I unpickle an iterator `iter` over an out-of-memory H5PyDataset which was pickled in the middle of epoch, I can't do `next(iter)`. In fact I have already traced the problem:...
Add the large MJSynth dataset of cropped words images for words recognition. The dataset was proposed in the paper Synthetic Data and Artificial Neural Networks for Natural Scene Text Recognition,...
Is it possible to use the H5PYDataset with different length sources? I am using vector features and image data as input but the number of distinct images is much less...
In `line 81`: `self.index += 1`, iterations are moving 1 frame at a time, even if `offset` was initialized `>1` `line 81` should be `self.index += self.offset` instead of `self.index...
