logo

Distributed training (#25)

* distributed training

* torchrun to start the training, but it quit directly

* fixed several distributed bugs. I am going to change the Dataset to map rather than IterableDataset since it does not work with DistributedSampler

* mode of dataset

* fix remove_contact bug

* distributed training is working.

whole brain affinity map training

lazily load blocks from image and label volumes inside the neuropil region.

boundary detection training code for 21_affinity_map_net

the boundary detection training is working.

semantic training for RNA granule

v0.1.0

semantic training

pre and post synapses detection training

for wasp ground truth datasets.

sample 3 post synapse detection

created a base class for datasets.

sample 3 tbar detection training

• smarter loading of ground truth volumes. We'll scan a directory and find the volumes out!

post synapse detection of sample 1

What's Changed

• use the complete Neuroglancer Precomputed Volume for unsupervised learning. by @jingpengw in #6
• add post_synapses dataset by @jingpengw in #8

New Contributors

• @jingpengw made their first contribution in #6

Full Changelog: v0.0.5...v0.0.6

t-bar detection of experiment 23

v0.0.5

fix perspective2d transform

tbar training of experiment 21

The result looks good.
I only found a few false-positive examples in small mitochondria with strip image artifact.