~/talk

Scaling Python for Performance: 40x Faster Image Processing with Smart designing

Talkat PyCon India 2025 · 2025-09-12..09-15· [python]
slides:none on file

Greedy rectangle-partitioning algorithm converts segmentation maps into exact non-overlapping rectangles and replaces deep recursion with a BFS task queue that dispatches tasks via multiprocessing and ProcessPoolExecutor, sharing big image buffers through a Python Manager to avoid copies. Hot inner loop is compiled with Numba JIT. On real wafer scans (≈8000×200k) runtime falls from 1100s → 18s and from 3500s → 28s; synthetic waffle tests drop 200s → 4s and worst-case weeks → 400s, giving up to 40× speedup.

AI-written summary · about

  1. 0:00Introduction and speaker intro
  2. 2:37Python scalability challenges
  3. 4:49Image segmentation problem
  4. 8:28Algorithm design approach
  5. 11:45GIL and multiprocessing limits
  6. 13:56Naive multiprocessing attempt
  7. 16:31Optimization strategies
  8. 19:01Queue-based multiprocessing
  9. 23:49Speedup techniques summary
  10. 24:51Performance results and benchmarks
  11. 26:18Key takeaways and conclusion
  12. 27:24Q&A session
~/search
type to search 6,518 conferences, 143,918 speakers and 192,133 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
← →
on a talk: 5 seconds back / forward
0–9
on a talk: jump to 0%–90% of the recording
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.