Code

Feat/orthoconv2d#51

Code

Closed

franckma31 wants to merge 25 commits intodevelopdeel-ai/deel-lip:developfrom feat/orthoconv2Ddeel-ai/deel-lip:feat/orthoconv2DCopy head branch name to clipboard

+2,408-28

Commits

Commits on Mar 14, 2022

• spectral_norm_conv and test_normalizers

  franckma31

  committed
  1ecec5f

  Copy full SHA for 1ecec5f

• add PadConv2D layers and test

  franckma31

  committed
  27d1fcd

  Copy full SHA for 27d1fcd

• PadConv2D add support for kernel_size as integer + test with several kernel and input size

  franckma31

  committed
  dd2b81d

  Copy full SHA for dd2b81d

• Add OrthoConv2D layer with spectral_norm_conv within the graph, and LorthRegul + unit tests

  franckma31

  committed
  60db98c

  Copy full SHA for 60db98c

• update _v initialization in spectral_norm with a tf.zeros since the first iteration will overwrite its value

  franckma31

  committed
  b364f28

  Copy full SHA for b364f28

• Add regularizer.py with Lorth regul

  Show description for 1512f5b
  franckma31

  committed
  1512f5b

  Copy full SHA for 1512f5b

• PEP-8 convention

  franckma31

  committed
  c776dbd

  Copy full SHA for c776dbd

• change file name and add utils methods for Singular Values computations + tests

  franckma31

  committed
  c07a013

  Copy full SHA for c07a013

• change file name and add utils methods for Singular Values computations + tests

  franckma31

  committed
  0737403

  Copy full SHA for 0737403

• add support for k_lip_tolerance_factor in test_layers

  franckma31

  committed
  282bb23

  Copy full SHA for 282bb23

• change function name compute_layer_sv and clean

  franckma31

  committed
  b85be94

  Copy full SHA for b85be94

• PEP-8 convention

  franckma31

  committed
  a4abf3e

  Copy full SHA for a4abf3e

• test compute_layers_sv for OrthoConv2D

  franckma31

  committed
  fa2fcaa

  Copy full SHA for fa2fcaa

• fix a test errors for tf2.8

  franckma31

  committed
  6546c3d

  Copy full SHA for 6546c3d

• bug fixed in test error: internal_input_shape for compute_output_shape

  franckma31

  committed
  658da78

  Copy full SHA for 658da78

• bug fix on FrobeniusConv2D::condense

  franckma31

  committed
  2703049

  Copy full SHA for 2703049

• FrobeniusConv2D vanilla_export bug correction + unit test

  franckma31

  committed
  4c30dc4

  Copy full SHA for 4c30dc4

• modify _power_iteration_conv for padding='same' and strides

  franckma31

  committed
  b4ef238

  Copy full SHA for b4ef238

• Correct _compute_lip_coef computation for spectralConv2D with strides; Stride parameter management in orthoConv2D

  franckma31

  committed
  3abbe0f

  Copy full SHA for 3abbe0f

• cleaning and naming convention

  franckma31

  committed
  525da45

  Copy full SHA for 525da45

• add unit test on strided convolutions

  franckma31

  committed
  3ded9d9

  Copy full SHA for 3ded9d9

• lint formatting

  franckma31

  committed
  e1aec2a

  Copy full SHA for e1aec2a

Commits on Mar 29, 2022

• compute_layer_sv.py: bug on compute_sv_add , and support for sequential models (without inputs) + snake case modifications

  franckma31

  committed
  dc7c036

  Copy full SHA for dc7c036

Commits on Apr 5, 2022

• orthoconv: changed camel case to snake case for attributes and methods

  thib-s

  committed
  6420318

  Copy full SHA for 6420318

• orthoconv: added missing documentation

  thib-s

  committed
  12d088a

  Copy full SHA for 12d088a