This software converts raw Timepix3 data (generated by the SPIDR readout) into HDF5 files. It can do clustering, and event localisation, and a number of corrections. On a decent 60 core workstation the code can parse about 1 Mhit/s data.
The green blocks are performed by tpx3HitParser while the blue blocks are done by tpx3EventViewer. The ToT correction file can be generated by these matlab scripts. The neural network model for event localisation is generated using these scrips.
git clone https://github.com/M4I-nanoscopy/tpx3HitParser.git
cd tpx3HitParser
Recommended way is to use a Python virtualenv, but this is optional.
A minimal version of Python 3.8 is needed (due to use of multiprocessing.shared_memory).
python3 -m venv tpx3-py3
source tpx3-py3/bin/activate
pip install pip --upgrade
pip install wheel
Install Python dependencies
pip install -r requirements.txt
For CNN event localisation Tensorflow 2.9 is being used. You will need a working NVIDIA CUDA installation. With CUDA Toolkit 11.2 and CUDNN 8.1.0.
For maximum speed in cluster finding, we're making use of a compiled library (written in Rust). Included is a pre-compiled version of this (Linux/amd64). If Python is unable to load this (i.e. you're on Windows) there is also a Python/numpy version of this same code. A warning is printed if this version is being used. This version is about 25% slower.
It should be possible to also compile the library for Windows using Rust/Cargo. No special dependencies are being used.
See the code at clusters/clfind.
$ ./tpx3HitParser --help
usage: tpx3HitParser.py [-h] [-c FILE] --raw FILE [-C] [-E] -o FILE [--overwrite] [--store_hits]
[--store_clusters] [--store_events] [--cluster_stats] [--hits_sort_toa 0/1]
[--event_sort_toa 0/1] [--freq_tot] [--cores N] [--max_hits N]
[--hits_remove_cross 0/1] [--hits_combine_chips 0/1]
[--hits_cross_extra_offset N] [--hits_tot_correct_file FILE]
[--hits_toa_phase_correction N] [--hits_tot_threshold N]
[--hits_correct_chip_edges 0/1] [--cluster_time_window N]
[--cluster_min_size N] [--cluster_max_size N] [--cluster_max_sum_tot N]
[--cluster_min_sum_tot N] [--cluster_chunk_size N] [--cluster_matrix_size N]
[--event_cnn_model FILE] [--event_cnn_tot_only 0/1]
[--event_correct_chip_edges 0/1] [-a A] [-v]
options:
-h, --help show this help message and exit
-c FILE, --config FILE Specify other config file (default: None)
-v, --verbose Verbose output (default: False)
input arguments:
--raw FILE Read raw .tpx3 (default: None)
parse arguments:
-C Parse clusters (default: False)
-E Parse events (default: False)
output arguments:
-o FILE, --output FILE Output HDF5 file (default: None)
--overwrite Overwrite existing HDF5 file (default: False)
--store_hits Store /hits in output file (default: False)
--store_clusters Store /clusters in output file (default: False)
--store_events Store /events in output file (default: False)
--cluster_stats Add the cluster sumToT and nHits to cluster_info or events
(default: False)
post processing:
--hits_sort_toa 0/1 Sort hit data on ToA (default: None)
--event_sort_toa 0/1 Sort event data on ToA (default: None)
--freq_tot Parse and store ToT frequency matrix (default: False)
constants:
--cores N Number of cores to use (default: 1)
--max_hits N Maximum number of hits to read (0: infinite) (default: 0)
--hits_remove_cross 0/1 Remove the middle border pixels between the chips (default: True)
--hits_combine_chips 0/1 Combine the chips to one matrix (default: True)
--hits_cross_extra_offset N Extra offset used for the cross pixels per chip when combining the
chips (default: 2)
--hits_tot_correct_file FILE ToT correction file, or 0 for no correction (default: 0)
--hits_toa_phase_correction N Apply ToA correction. 0=None, 1=Maastricht-Pll30, 2=Basel-Pll30,
3=Pll94 (default: 0)
--hits_tot_threshold N Below this ToT threshold hits are not stored (default: 5)
--hits_correct_chip_edges 0/1 Correct chip edges for hits (not applied on data used for
clustering) (default: None)
--cluster_time_window N Maximum time interval between individual hits to cluster them (in
fine ToA values=1.56ns) (default: 50)
--cluster_min_size N Minimum cluster size (default: 2)
--cluster_max_size N Maximum cluster size (default: 10)
--cluster_max_sum_tot N Maximum cluster sum tot (default: 400)
--cluster_min_sum_tot N Minimum cluster sum tot (default: 200)
--cluster_chunk_size N Number of hits to consider at once for clustering. (default:
10000)
--cluster_matrix_size N Size of the resulting cluster matrix (default: 10)
--event_cnn_model FILE CNN model to use for event localisation (default: model-200kv-
tottoa.h5)
--event_cnn_tot_only 0/1 The specified CNN model uses ToT only (default: False)
--event_correct_chip_edges 0/1 Correct chip edge for events (default: None)
-a A, --algorithm A Event localisation algorithm to use (default: centroid)
All options are command line options. Defaults for constants are specified in a separate config file default.cfg. You can either
edit this file or make your own and specify this with the --config option.
Please consider citing either or both the Zenodo deposit of this code and our paper:
- van Schayck, J. Paul. (2020). M4I-nanoscopy/tpx3EventViewer. Zenodo. https://doi.org/10.5281/zenodo.3693990
- Schayck, J. P. van, Genderen, E. van, Maddox, E., Roussel, L., Boulanger, H., Fröjdh, E., Abrahams, J.-P., Peters, P. J. & Ravelli, R. B. G. (2020). Sub-pixel electron detection using a convolutional neural network. Ultramicroscopy, 218, 113091. https://doi.org/10.1016/j.ultramic.2020.113091
- J Paul van Schayck, Yue Zhang, Kèvin Knoops, Peter J Peters, Raimond B G Ravelli, Integration of an Event-driven Timepix3 Hybrid Pixel Detector into a Cryo-EM Workflow, Microscopy and Microanalysis, Volume 29, Issue 1, February 2023, Pages 352–363, https://doi.org/10.1093/micmic/ozac009
(c) Maastricht University
MIT License
- Paul van Schayck ([email protected])
- Raimond Ravelli ([email protected]) (corresponding)