π Geometric Computer Vision Library for Spatial AI
Python artificial-intelligencecomputer-visiondeep-learninghacktoberfestimage-processingmachine-learningneural-networkpythonpytorchroboticsspatial-ai
Commits on main
docs(augmentation): adopt choose-then-use docstrings and catalogs (#6560)
60b0618 Dmytro Mishkin committed 13 days ago
fix(feature): detect a float16 image in float32 in ScaleSpaceDetector (#6606)
d8a6850 Dmytro Mishkin committed 13 days ago
fix(geometry): build the warps' matrices and sampling grids outside a caller's autocast (#6596)
4dd7f94 Dmytro Mishkin committed 13 days ago
fix(geometry): record list padding per box so writes cannot make it stale (#6595)
9578ff6 Dmytro Mishkin committed 13 days ago
fix(enhance): keep the derivative at exact 0 and 1 in the [0, 1] intensity clamps (#6594)
7200ba9 Dmytro Mishkin committed 13 days ago
fix(augmentation): run auto-augment operations with a magnitude after .to(device) (#6593)
8a48e19 Dmytro Mishkin committed 13 days ago
fix(core): avoid the faulty Windows SDPA kernel in all attention callers (#6613)
764c4d8 Dmytro Mishkin committed 13 days ago
docs: consolidate correctness policy and contribution guides (#6611)
aa3326c Dmytro Mishkin committed 13 days ago
fix(feature): compute half-precision dominant orientation in float32 (#6590)
730ef2e Dmytro Mishkin committed 13 days ago
test: fix CUDA-only torch.compile test contracts (resized-crop nearest oracle, fit_line, fit_plane, Keypoints) (#6586)
b914d37 Dmytro Mishkin committed 14 days ago
test: give CUDA gradchecks a 1e-10 nondet_tol for run-to-run rounding of the default kernels (#6587)
7645e81 Dmytro Mishkin committed 14 days ago
Fix Lie group shape validation and quaternion coefficient handling (#6513)
760326f Bhavya Kumar committed 14 days ago
fix(geometry): build undistort_image's half-precision map in float32 (#6577)
b73e4f4 Viranshul-P committed 14 days ago
fix(color): resample a ColorMap palette from its first colour to its last (#6557)
43ad6c2 Saksham Sharma committed 14 days ago
docs(enhance): solarize takes a float or one-element tensor, not an int (#6551)
99d42f4 Saksham Sharma committed 14 days ago
Preserve KMeans accumulation precision under autocast (#6597)
f50e332 A Aswanth Raj committed 14 days ago
fix(geometry): accept a differing cx in StereoCamera, as OpenCV's stereoRectify produces (#6558)
6609c75 Prateekshit Jaiswal committed 14 days ago
fix(geometry): keep a singleton box axis when indexing an unbatched box with a scalar (#6484)
42b42c4 niukanen1 committed 14 days ago
fix(dev): make the CUDA install task use the locked PyTorch (#6599)
5310f25 Dmytro Mishkin committed 14 days ago
fix(geometry): return one point per plane from ParametrizedLine.intersect (#6589)
28bf2be Dmytro Mishkin committed 14 days ago
fix(geometry): keep the polynomial solver's tables out of each call's device and dtype (#6588)
3aa458b Dmytro Mishkin committed 14 days ago
test: give CUDA half batch-vs-single comparisons a half tolerance (LoFTR, SmallSR, LightGlue) (#6584)
2620e13 Dmytro Mishkin committed 15 days ago
fix(filters): bin Otsu pixels with a correctly rounded division under CUDA torch.compile (#6585)
1c84c52 Dmytro Mishkin committed 15 days ago
docs: replace the historical CUDA test counts with a full CUDA run of e4a444e9b (#6579)
5a37e7a Dmytro Mishkin committed 15 days ago
fix(geometry): optimize float16 and bfloat16 ImageRegistrator models in float32 (#6568)
e4a444e Dmytro Mishkin committed 15 days ago
fix(feature): set the patch convolution layout in scripted SOSNet, TFeat and HyNet as in eager (#6567)
5d23ba2 Dmytro Mishkin committed 15 days ago
fix(filters): transform float16 and bfloat16 inputs of fft_conv in float32 on every device (#6564)
5ef0925 Dmytro Mishkin committed 15 days ago
fix(geometry): run float16 and bfloat16 warps in float32 on every device (#6563)
02d3a70 Dmytro Mishkin committed 15 days ago
fix(augmentation): deep-copy and pickle augmentations with learnable ranges (#6541)
77c088e Dmytro Mishkin committed 15 days ago