DeepCAD is a self-supervised deep-learning method for spatiotemporal enhancement of calcium imaging data that does not require any high signal-to-noise ratio (SNR) observations. DeepCAD suppresses detection noise and improves the SNR more than tenfold, which reinforces the accuracy of neuron extraction and spike inference and facilitates the functional analysis of neural circuits.
Allocate an interactive session and run the program.
Sample session on a GPU node:
[user@biowulf ~]$ sinteractive --mem=48g -c4 --gres=gpu:k80:1,lscratch:10 [user@cn4219 ~]$ module load deepcad [+] Loading singularity 3.8.4 on cn4219 [+] Loading cuDNN 7.0 libraries... [+] Loading CUDA Toolkit 9.0.176 ... [+] Loading deepcad 20210826Basic usage:
[user@cn4219 ~]$ main.py -h
...
usage: main.py [-h] [--img_h IMG_H] [--img_w IMG_W] [--img_s IMG_S]
[--img_c IMG_C] [--gap_h GAP_H] [--gap_w GAP_W] [--gap_s GAP_S]
[--normalize_factor NORMALIZE_FACTOR]
[--train_epochs TRAIN_EPOCHS] [--datasets_path DATASETS_PATH]
[--datasets_folder DATASETS_FOLDER]
[--DeepCAD_model_folder DEEPCAD_MODEL_FOLDER]
[--results_folder RESULTS_FOLDER] [--GPU GPU]
[--is_training IS_TRAINING] [--lr LR]
[--train_datasets_size TRAIN_DATASETS_SIZE]
[--select_img_num SELECT_IMG_NUM]
optional arguments:
-h, --help show this help message and exit
--img_h IMG_H the height of patch stack
--img_w IMG_W the width of patch stack
--img_s IMG_S the image number of patch stack
--img_c IMG_C the channel of image
--gap_h GAP_H the height of patch gap
--gap_w GAP_W the width of patch gap
--gap_s GAP_S the image number of patch gap
--normalize_factor NORMALIZE_FACTOR
Image normalization factor
--train_epochs TRAIN_EPOCHS
train epochs
--datasets_path DATASETS_PATH
the name of your project
--datasets_folder DATASETS_FOLDER
the folders for datasets
--DeepCAD_model_folder DEEPCAD_MODEL_FOLDER
the folders for DeepCAD(pb) model
--results_folder RESULTS_FOLDER
the folders for results
--GPU GPU the index of GPU you will use for computation
--is_training IS_TRAINING
train or test
--lr LR initial learning rate
--train_datasets_size TRAIN_DATASETS_SIZE
actions: train or predict
--select_img_num SELECT_IMG_NUM
actions: train or predict
Create an input data folder and run DeepCAD on the data in that folder:
[user@cn4219 ~]$ mkdir -p datasets/my_data
[user@cn4219 ~]$ cp $DEEPCAD_DATA/* datasets/my_data
[user@cn4219 ~]$ main.py --GPU 0 --img_h 64 --img_w 64 --img_s 320 --train_epochs 30 --datasets_folder my_data --normalize_factor 1 --lr 0.00005 --train_datasets_size 100 &
list(os.walk(im_folder, topdown=False)) -----> [('datasets//my_data', [], ['ForTestPlugin_256x256x1000.tif'])]
stack_num -----> 1
TiffPage 0: TypeError: read_bytes() missing 3 required positional arguments: 'dtype', 'count', and 'offsetsize'
whole_w -----> 256
whole_h -----> 256
whole_s -----> 1000
w_num -----> 4
h_num -----> 4
s_num -----> 7
gap_s -----> 60
2021-11-19 15:37:52.507610: I tensorflow/core/platform/cpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
2021-11-19 15:37:52.507957: E tensorflow/stream_executor/cuda/cuda_driver.cc:406] failed call to cuInit: CUresult(-1)
2021-11-19 15:37:52.508015: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:158] retrieving CUDA diagnostic information for host: cn4219
2021-11-19 15:37:52.508042: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:165] hostname: cn4219
2021-11-19 15:37:52.508197: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:189] libcuda reported version is: Not found: was unable to find libcuda.so DSO loaded into this program
2021-11-19 15:37:52.508245: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:369] driver version file contents: """NVRM version: NVIDIA UNIX x86_64 Kernel Module 470.82.01 Wed Oct 27 21:21:55 UTC 2021
GCC version: gcc version 4.8.5 20150623 (Red Hat 4.8.5-28) (GCC)
"""
2021-11-19 15:37:52.508289: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:193] kernel reported version is: 470.82.1
--- Epoch 0 --- Step 0 / 112 --- L1_loss 382.85928 --- L2_loss 362411.6 --- Time 1013.0063416957855
train_input ---> 14906.737 ---> 0.0
output_img ---> 1.1468295 ---> -0.42850912
/usr/local/apps/deepcad/20210826/src/main.py:89: UserWarning: results//unet3d_my_data_20211119-1537//0_0_my_data_ForTestPlugin_256x256x1000_x1_y2_z3_output.tif is a low contrast image
io.imsave(result_name, output_img.transpose(2,0,1))
/usr/local/apps/deepcad/20210826/src/main.py:90: UserWarning: results//unet3d_my_data_20211119-1537//0_0_my_data_ForTestPlugin_256x256x1000_x1_y2_z3_noise1.tif is a low contrast image
io.imsave(noise_img1_name, train_input.transpose(2,0,1))
/usr/local/apps/deepcad/20210826/src/main.py:91: UserWarning: results//unet3d_my_data_20211119-1537//0_0_my_data_ForTestPlugin_256x256x1000_x1_y2_z3_noise2.tif is a low contrast image
io.imsave(noise_img2_name, train_GT.transpose(2,0,1))
...
[user@cn4219 ~]$ nvidia-smi
Fri Nov 19 15:37:53 2021
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.82.01 Driver Version: 470.82.01 CUDA Version: 11.4 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 Tesla K80 On | 00000000:8A:00.0 Off | 0 |
| N/A 79C P0 140W / 149W | 10955MiB / 11441MiB | 100% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| 0 N/A N/A 21956 C ...a/envs/deepcad/bin/python 10950MiB |
+-----------------------------------------------------------------------------+
...
--- Epoch 1 --- Step 0 / 112 --- L1_loss 335.09457 --- L2_loss 224135.25 --- Time 1144.084424495697
train_input ---> 6691.333 ---> 0.0
output_img ---> 1.9556417 ---> 0.14389074
/usr/local/apps/deepcad/20210826/src/main.py:89: UserWarning: results//unet3d_my_data_20211120-0843//1_0_my_data_ForTestPlugin_256x256x1000_x0_y3_z4_output.tif is a low contrast image
io.imsave(result_name, output_img.transpose(2,0,1))
/usr/local/apps/deepcad/20210826/src/main.py:90: UserWarning: results//unet3d_my_data_20211120-0843//1_0_my_data_ForTestPlugin_256x256x1000_x0_y3_z4_noise1.tif is a low contrast image
io.imsave(noise_img1_name, train_input.transpose(2,0,1))
/usr/local/apps/deepcad/20210826/src/main.py:91: UserWarning: results//unet3d_my_data_20211120-0843//1_0_my_data_ForTestPlugin_256x256x1000_x0_y3_z4_noise2.tif is a low contrast image
io.imsave(noise_img2_name, train_GT.transpose(2,0,1))
...
--- Epoch 20 --- Step 0 / 112 --- L1_loss 281.3424 --- L2_loss 174512.05 --- Time 21978.92845606804
train_input ---> 7301.258 ---> 0.0
output_img ---> 25.690779 ---> 6.511296
/usr/local/apps/deepcad/20210826/src/main.py:89: UserWarning: results//unet3d_my_data_20211120-0843//20_0_my_data_ForTestPlugin_256x256x1000_x1_y0_z3_output.tif is a low contrast image
io.imsave(result_name, output_img.transpose(2,0,1))
/usr/local/apps/deepcad/20210826/src/main.py:90: UserWarning: results//unet3d_my_data_20211120-0843//20_0_my_data_ForTestPlugin_256x256x1000_x1_y0_z3_noise1.tif is a low contrast image
io.imsave(noise_img1_name, train_input.transpose(2,0,1))
/usr/local/apps/deepcad/20210826/src/main.py:91: UserWarning: results//unet3d_my_data_20211120-0843//20_0_my_data_ForTestPlugin_256x256x1000_x1_y0_z3_noise2.tif is a low contrast image
io.imsave(noise_img2_name, train_GT.transpose(2,0,1))
...
Create a batch input file (e.g. deepcad.sh). For example:
#!/bin/bash set -e module load deepcad mddir datasets/my_data cp $DEEPCAD_DATA/* datasets/my_data main.py --GPU 0 --img_h 64 --img_w 64 --img_s 320 --train_epochs 30 --datasets_folder my_data --normalize_factor 1 --lr 0.00005 --train_datasets_size 100
Submit this job using the Slurm sbatch command.
sbatch deepcad.sh