Analyzing 3D objects with power of Deep Learning and Cython
summary
Claims a working Python module PySparseConvNet that marries C++/CUDA kernels to Python via Cython and specific .pxd/.pyx patterns, with attention to explicit init/dealloc and type mapping. Implementation compiles CUDA objects in parallel and uses a Google Sheets–backed hyperparameter controller. Evaluated on 40 classes with roughly 12,000 samples, getting accuracy comparable to multi-view and dense baselines while using far less computation by operating only on sparse voxels.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 2:063D shape representation challenges
- 5:27pi sparse canet purpose
- 8:39Sifan notation Python-to-C++
- 14:10Compilation process
- 18:55Shape retrieval metric learning
- 22:08Sparse CNN architecture
- 27:58Performance vs state-of-the-art
- 29:24Hyperparameter optimization
- 31:55Q&A
related talks
30mCuPy: A NumPy-compatible Library for GPU
57mDeep Learning with C++
55mA Tour of Deep Learning With C++
44mIntro to ConvNets
01h 02mExploit your GPU power with PyCUDA (and friends)
02h 28mGPU Accelerated Python
30mCUDA in Your Python: Effective Parallel Programming on the GPU
02h 55mGPU Accelerated Python w: CUDA