DocGarbanzo
DocGarbanzo
Hi @JacobGlennAyers - If you say faulty images, do you mean a corrupted `.jpg` file, or is the whole file missing from the `images` directory?
I have never seen this before. From your listing below it looks like there is a continuous set of images (..., 419, 420, 421,...) which failed to write. The timestamp...
Ok, so if you want to filter corrupted images at the training state, I would not recommend to manually edit the catalog files and their manifest files in order not...
> @DocGarbanzo By corrupt indexes, do you mean with respect to their order in the "images" folder nested in the "data" folder? Not necessarily the order, I meant the index...
@JacobGlennAyers - the most robust solution would be to try loading the image file. You can do this through building the generator or list of `TubRecords` from the Iterable `Tub`...
Hi @blaughlin - your code can't be merged, you have to rebase first.
The Travis failure is w.r.t the training convergence which we have already scheduled to take out of travis, everything else passes. Looks good to me otherwise.
@wallarug - is this still a relevant PR? This is related to #583.
@CarloHan - can you let us know if you solve the issue?
Closed due to inactivity.