TABLE 3. Memory and FLOP comparison.
Comparison of memory consumption and FLOPs for five unrolled iterations of the ADMM block in Le-ADMM (full and 4X downsampled versions) and the trainable inversion stage of our proposed FlatNet-gen.
| Method | Memory (in MB) | Computation (in MFLOP) |
|---|---|---|
| Le-ADMM-Full | 6300 | 1290 |
| Le-ADMM-Downsampled | 1000 | 65 |
| FlatNet-gen | 990 | 53 |