Synaptic Cleft Detection

This tutorial provides step-by-step guidance for synaptic cleft detection with CREMI benchmark datasets. We consider the task as a semantic segmentation task and predict the synapse pixels with encoder-decoder ConvNets similar to the models used in affinity prediction in neuron segmentation. The evaluation of the synapse detection results is based on the F1 score and average distance. See CREMI metrics for more details.

Note

We preform re-alignment of the original CREMI image stacks and also remove the crack artifacts. Please reverse the alignment before submitting the test prediction to the CREMI challenge.

All the scripts needed for this tutorial can be found at pytorch_connectomics/scripts/. Need to pass the argument --task 1 when executing the train.py and test.py scripts. The pytorch dataset class of synapses is torch_connectomics.data.dataset.SynapseDataset.

1. Get the dataset:

   1. Download the dataset from our server:

      wget http://hp06.mindhackers.org/rhoana_product/dataset/cremi.zip
      

   For description of the data please check this page.

2. Run the training script. The training and inference script can take a list of volumes and conduct training/inference at the same time.

   $ source activate py3_torch
   $ python -u train.py -i /path/to/CREMI/ \
     -din image/im_A_v2_200.h5@image/im_B_v2_200.h5@image/im_C_v2_200.h5 \
     -dln gt-syn/syn_A_v2_200.h5@gt-syn/syn_B_v2_200.h5@gt-syn/syn_C_v2_200.h5 \
     -o outputs/unetv0_syn -lr 1e-03 --iteration-total 50000 --iteration-save 5000 \
     -mi 8,256,256 -ma unet_residual_3d -moc 1 \
     -to 0 -lo 1 -wo 1 -g 4 -c 4 -b 8
   

   • data: i/o/din/dln (input folder/output folder/train volume/train label)

   • optimization: lr/iteration-total/iteration-save (learning rate/total #iterations/#iterations to save)

   • model: mi/ma/moc (input size/architecture/#output channel)

   • loss: to/lo/wo (target option/loss option/weight option)

   • system: g/c/b (#GPU/#CPU/batch size)

3. Visualize the training progress:

   $ tensorboard --logdir runs
   

4. Run inference on image volumes:

   $ CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -u test.py -i /path/to/CREMI/ \
     -din image/im_A_v2_200.h5@image/im_B_v2_200.h5@image/im_C_v2_200.h5 \
     -o outputs/unetv0_syn/result -mi 8,256,256 -ma unet_residual_3d -moc 1 \
     -g 4 -c 4 -b 32
     -mpt outputs/unetv0_syn/volume_49999.pth -mpi 49999 -dp 8,64,64 -tam mean -tan 4
   

   • pre-train model: mpt/mpi (model path/iteration number)

   • test configuration: dp/tam/tan (data padding/augmentation mode/augmentation number)