tihbe/python-ebdataset

An event based dataset loader under one common python API.

python-ebdataset's Issues

Typo in readme

There is a typo in usage section:

dt = 1*ms
loader = NMnist(path, is_train=True, transform=ToDense(dt))

The argument transform should be transforms

Reading NMNIST Dataset

Hi,

I am trying to read the NMNIST Dataset by using your code.

I have files like "00001.bin", in short, binary format.

I have got this error:

I don't know how to solve this error?

Can you please help me out what's the problem?

Thanks

Possible bug with ibm gesture

Here is minimal working example:

from ebdataset.vision import IBMGesture, H5IBMGesture
from ebdataset.vision.transforms import ToDense
from quantities import ms
H5IBMGesture.convert(DVS_GESTURE_TAR, H5_FILE)
data = H5IBMGesture(H5_FILE)
spike_train = data[0][0]
convert = ToDense(10*ms)
dense = convert(spike_train)

Error message:

index 26880 is out of bounds for dimension 0 with size 128
  File "/home/myhomefolder/pytorch_env/lib/python3.6/site-packages/ebdataset/vision/transforms.py", line 72, in __call__
    ] = 1

I found that the error is because of x and y in dvs spike train, please see below:

spike_train.x
array([   49,    82,  7169, ...,     0,     0, 21248], dtype=uint16)
spike_train.x.max()
65531

dvs gesture size is 128x128, however, I can find values that are greater than 128 in both spike_train.x and spike_train.y

My environment:
torch==1.2.0
numpy==1.17.4
torchvision==0.4.0
h5py==3.1.0
quantities==0.12.4