Skip to content

gigantocypris/cryo-segment

Repository files navigation

cryo-segment

Upload file to Perlmutter: scp /Users/VGanapati/Dropbox/XFEL_LBNL/cryo_ET/Liposome_Data/mon_t1_trimmed.rec.nad [email protected]:/global/cfs/cdirs/m3562/users/vidyagan/output-cryo-segment

Log on to Perlmutter

cd /global/cfs/cdirs/m3562/users/vidyagan git clone https://github.com/vganapati/cryo-segment.git

module load python module load pytorch/2.0.1

module load pytorch/2.1.0-cu12

conda create -n cryo conda activate cryo

pip install -U cryoet-data-portal pip install -U mrcfile

Opening MRC files https://mrcfile.readthedocs.io/en/stable/usage_guide.html

Data Portal GitHub: https://github.com/chanzuckerberg/cryoet-data-portal/tree/main

Getting Started after Setup:

Log on to Perlmutter "Take" data:

cd /global/cfs/cdirs/m3562/users/vidyagan/output-cryo-segment take -u nksauter emd_10439.map.gz

module load python module load pytorch/2.0.1

module load pytorch/2.1.0-cu12

conda activate cryo cd /global/cfs/cdirs/m3562/users/vidyagan/cryo-segment

Running the Zhou et al pipeline:

mkdir checkpoint mkdir ciro cp Zhou_2023/UNET/data/files/*.txt ciro/ python Zhou_2023/UNET/data/createcirosimnewpart.py (make sure line 118 is "dir_name = 'ciro'" creates ciro folder in working directory. copy train.txt, test.txt and labels.txt to this folder, make a "checkpoint" folder in the working directory) salloc --nodes 1 --qos interactive --time 01:00:00 --constraint gpu --gpus 1 --account=m3562 python Zhou_2023/UNET/train.py -e NUM_EPOCHS (try 50 epochs to start, 500 epochs for production)

Creating the test images: Change Line 118 of "Zhou_2023/UNET/data/createcirosimnewpart.py" to dir_name='val' and run the code to generate the test data. (originally, dir_name = 'ciro')

python Zhou_2023/UNET/data/createcirosimnewpart.py

Following the README directions:

Run the code "python Zhou_2023/UNET/visunet.py" to test result on one images. Change the image file by change the code:

Line 31: image_path = 'data/val/200_3_data.png'

(created "result/vis" folder in working directory) check the results in "result/vis" in the working directory

python Zhou_2023/UNET/visunet.py

3.3 test a group model on a group images.

Run the code "visgroupunet.py" to test result by group. It will test allChange model in "checkpoint/", Change the image file by change the code:

Line 53: image_list=['100_0_data.png','200_0_data.png','300_0_data.png','400_0_data.png','500_0_data.png','600_0_data.png','700_0_data.png','800_0_data.png','900_0_data.png','1000_0_data.png']

check the results in "result" in working directory

python Zhou_2023/UNET/visgroupunet.py

3.4 test model on one slice of real membrane files.

Run the code "visunetmen.py",change the configuration as below.

python Zhou_2023/UNET/visunetmen.py

Need MRC file: /global/cfs/cdirs/m3562/users/vidyagan/output-cryo-segment/dataset_10010_TE2 /global/cfs/cdirs/m3562/users/vidyagan/output-cryo-segment/dataset_10010_TE13

Line 31 : model_path_root='checkpoint/' #models directory. Line 32 : image_path = '/global/cfs/cdirs/m3562/users/vidyagan/output-cryo-segment/mon_t1_trimmed.rec.nad' #which file used for test. Line 36 : model_path=model_path='checkpoint/checkpoint_450.pth' #model_list[0] #which model used for test. Line 38 : index=100 #which slice used for test.

python Zhou_2023/UNET/visunetmen_mrc.py 3.5 test model on all slices of real membrane files. Run the code "visunetmen_mrc.py",change the configuration as below.

Line 31 : model_path_root='checkpoint/' #models directory. Line 32 : image_path = '/global/cfs/cdirs/m3562/users/vidyagan/output-cryo-segment/mon_t1_trimmed.rec.nad' #which file used for test. Line 36 : model_path=model_path='checkpoint/checkpoint_450.pth' #model_list[0] #which model used for test.

Results will be saved in the same folder as mon_t1_trimmed.rec.nad (e.g. output-cryo-segment) Next steps:

RL

python Zhou_2023/RL/mainall.py python Zhou_2023/RL/main7.py

RegionMerge

GP3D

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published