enijkamp

from PIL import Image
import glob

for p in glob.glob('*.png'):
    with open(p, 'rb') as f:
        img = Image.open(f)
        try:
            img.convert('RGB')
        except:
            print(p)

enijkamp

class AddCoords(nn.Module):

    def __init__(self, with_r=False):
        super().__init__()
        self.with_r = with_r

    def forward(self, input_tensor):
        batch_size, _, x_dim, y_dim = input_tensor.size()

        xx_channel = torch.arange(x_dim).repeat(1, y_dim, 1)

enijkamp

class AddCoords(nn.Module):

    def __init__(self, with_r=False):
        super().__init__()
        self.with_r = with_r

    def forward(self, input_tensor):
        batch_size, _, x_dim, y_dim = input_tensor.size()

        xx_channel = torch.arange(x_dim).repeat(1, y_dim, 1)

enijkamp

import os

import pandas as pd
import numpy as np
import argparse

import torch
import torchvision

import models

enijkamp

  0 cs1=  0.2207 cs2=  1.3196 c1=  0.9443 c2=  0.6621
  1 cs1=  0.2215 cs2=  1.3286 c1=  0.9683 c2=  0.6934
  2 cs1=  0.2242 cs2=  1.3213 c1=  0.9744 c2=  0.7188
  3 cs1=  0.2252 cs2=  1.3243 c1=  0.9749 c2=  0.7258
  4 cs1=  0.2258 cs2=  1.3321 c1=  0.9778 c2=  0.7244
  5 cs1=  0.2255 cs2=  1.3320 c1=  0.9788 c2=  0.7263
  6 cs1=  0.2247 cs2=  1.3337 c1=  0.9780 c2=  0.7319
  7 cs1=  0.2252 cs2=  1.3411 c1=  0.9788 c2=  0.7273
  8 cs1=  0.2257 cs2=  1.3427 c1=  0.9800 c2=  0.7319
  9 cs1=  0.2249 cs2=  1.3394 c1=  0.9802 c2=  0.7341

enijkamp

  0 cs1=  0.2292 cs2=  1.3196 cs1_l=  0.2292 cs2_l=  1.3196 c1=  0.9407 c2=  0.6621
  1 cs1=  0.2308 cs2=  1.3286 cs1_l=  0.1636 cs2_l=  1.1018 c1=  0.9634 c2=  0.6934
  2 cs1=  0.2333 cs2=  1.3213 cs1_l=  0.1473 cs2_l=  1.0273 c1=  0.9688 c2=  0.7188
  3 cs1=  0.2343 cs2=  1.3243 cs1_l=  0.1395 cs2_l=  0.9930 c1=  0.9692 c2=  0.7258
  4 cs1=  0.2335 cs2=  1.3321 cs1_l=  0.1339 cs2_l=  0.9786 c1=  0.9734 c2=  0.7244
  5 cs1=  0.2314 cs2=  1.3320 cs1_l=  0.1292 cs2_l=  0.9649 c1=  0.9751 c2=  0.7263
  6 cs1=  0.2318 cs2=  1.3337 cs1_l=  0.1272 cs2_l=  0.9550 c1=  0.9763 c2=  0.7319
  7 cs1=  0.2313 cs2=  1.3411 cs1_l=  0.1246 cs2_l=  0.9546 c1=  0.9788 c2=  0.7273
  8 cs1=  0.2301 cs2=  1.3427 cs1_l=  0.1224 cs2_l=  0.9503 c1=  0.9790 c2=  0.7319
  9 cs1=  0.2287 cs2=  1.3394 cs1_l=  0.1203 cs2_l=  0.9418 c1=  0.9805 c2=  0.7341

enijkamp

loss1=    0.1201 lr=  0.000100
2019-01-14 07:50:29,196 : epoch 153  , step 277  : loss1=    0.1204 lr=  0.000100
2019-01-14 07:50:29,358 : epoch 153  , step 278  : loss1=    0.1205 lr=  0.000100
2019-01-14 07:50:29,607 : epoch 153  , step 279  : loss1=    0.1201 lr=  0.000100
2019-01-14 07:50:29,776 : epoch 153  , step 280  : loss1=    0.1208 lr=  0.000100
2019-01-14 07:50:29,898 : epoch 153  , step 281  : loss1=    0.1199 lr=  0.000100
2019-01-14 07:50:30,070 : epoch 153  , step 282  : loss1=    0.1207 lr=  0.000100
2019-01-14 07:50:30,184 : epoch 153  , step 283  : loss1=    0.1203 lr=  0.000100
2019-01-14 07:50:30,315 : epoch 153  , step 284  : loss1=    0.1203 lr=  0.000100
2019-01-14 07:50:30,460 : epoch 153  , step 285  : loss1=    0.1208 lr=  0.000100

enijkamp

2019-01-20 01:54:49,042 : epoch 210  , step 385   loss0=    0.0000 loss1=   29.9530 loss2=    0.0000 lr=  0.001000
2019-01-20 01:54:49,132 : epoch 210  , step 386   loss0=    0.0000 loss1=   30.0526 loss2=    0.0000 lr=  0.001000
2019-01-20 01:54:49,243 : epoch 210  , step 387   loss0=    0.0000 loss1=   30.1210 loss2=    0.0000 lr=  0.001000
2019-01-20 01:54:49,327 : epoch 210  , step 388   loss0=    0.0000 loss1=   30.0791 loss2=    0.0000 lr=  0.001000
2019-01-20 01:54:49,421 : epoch 210  , step 389   loss0=    0.0000 loss1=   29.9636 loss2=    0.0000 lr=  0.001000
2019-01-20 01:54:49,514 : epoch 210  , step 390   loss0=    0.0000 loss1=   29.8071 loss2=    0.0000 lr=  0.001000
2019-01-20 01:55:04,683 : epoch=210 n=10 acc=0.7752 (target)
2019-01-20 01:55:23,738 : >>  epoch 210  : average_loss=  0.7859, average_accuracy=73.131%,  mean_weight_std=0.09486, mean_normalized_std=11.0705  mean_dist=  36.55
2019-01-20 01:55:23,744 : >>  epoch 210  : median_loss=  0.7838, median_accuracy=73.045%,  median_weight_std

enijkamp

2019-01-20 02:22:59,985 : >>  epoch 20   : dist pair-wise (target)
2019-01-20 02:22:59,987 : 
[[ 0.    11.072 10.757 11.233 11.483 10.696 11.791 11.471 12.791 11.896]
 [11.072  0.     9.685 10.94  10.753 10.934 10.456 11.523 11.475 11.205]
 [10.757  9.685  0.    10.439 10.51  11.268 10.206 10.785 11.371 10.588]
 [11.233 10.94  10.439  0.    10.959 11.421 11.273 11.323 11.763 11.446]
 [11.483 10.753 10.51  10.959  0.    11.945 10.611 11.584 10.717 10.703]
 [10.696 10.934 11.268 11.421 11.945  0.    12.167 11.694 12.71  12.403]
 [11.791 10.456 10.206 11.273 10.611 12.167  0.    11.947 10.735 10.685]
 [11.471 11.523 10.785 11.323 11.584 11.694 11.947  0.    12.685 12.144]

enijkamp

2019-01-20 23:27:25,717 : set device 1 out of 4 cuda devices
2019-01-20 23:27:25,717 : Namespace(checkpoint=None, data_path='./data', dataset='CIFAR10', device=1, e_increasing_layer_size=False, e_layer_size=200, e_number_of_hidden_layers=3, epochs=100000000000000000000, gamma=1.0, initial_weights=None, lamBda=0.0, lr_init=0.0001, model='LeNet', momentum=0.0, num_workers=4, resume_checkpoint=False, save_checkpoint=True, seed=2092, train_log_freq=1, w_batch_size=10, weight_sharing=False, wg_number_of_hidden_layers=2, with_adam=True, with_lr_plateau_schedule=False, with_norm=2, with_residual=False, x_batch_size=512, z_batch_size=10, z_dim=100, z_std=1.0)
2019-01-20 23:27:30,267 : {'weight_sharing': False, 'input_noise_size': 100, 'with_layernorm': False, 'with_batchnorm': True, 'e_increasing_layer_size': False, 'e_number_of_hidden_layers': 3, 'e_layer_size': 200, 'wg_number_of_hidden_layers': 2, 'wg_hidden_layer_size_formula': <function train.<locals>.wg_hidden_layer_size_formula at 0x7f7ed2bcdb70>, 'code_size_f