Table 2. Total time to reconstruct the datasets listed in Table 1 ▸ using TomocuPy’s FourierRec and LpRec methods with FP16 and FP32 precision, TomocuPy’s LineRec with FP32 precision, and the TomoPy-cli package where all pre-processing steps and the Gridrec method for reconstruction are executed in FP32 precision.
Raw data (16-bit) and reconstructed volumes (16- or 32-bit) both have sizes N × N × N, assuming the number of projection angles is equal to N. Reconstruction time includes all parts of the processing pipeline (reading HDF5 data chunks from NVMe SSD, writing reconstructed tiff files to NVMe SSD, CPU-GPU transfers, and all CPU/GPU computations).
| Test dataset | 1 | 2 | 3 | 4 | 5 | 6 |
|---|---|---|---|---|---|---|
| FourierRec, 16-bit |
s |
s |
s |
s |
s |
s |
| FourierRec, 32-bit |
s |
s |
s |
s |
s |
– |
| LpRec, 16-bit |
s |
s |
s |
s |
s |
s |
| LpRec, 32-bit |
s |
s |
s |
s |
s |
– |
| LineRec |
s |
s |
s |
s |
s |
– |
| TomoPy Gridrec, 32-bit |
s |
s |
s |
s |
s |
s†
|
†
Estimated by using a lower number of chunks.