KaneAI - World’s First E2E Software Testing Agent.Book a Demo Join TestMu Conference

How to use base_resize method in toolium

Best Python code snippet using toolium_python

optimization_l2.py

Source:optimization_l2.py

1import json2import time3import torch4from distutils.dir_util import copy_tree5from DC.common import vgg_normalize, vgg_renormalize, d2_distance_matrix, select_distance_matrix_idx, \6    sort_by_z, custom_vgg, remove_invisible_z7from DC.compositing import composite_layers8from DC.constants import *9from DC.loss import loss_fn_l210from DC.utils import load_element, load_base_pattern, init_element_pos, render_constants, make_cuda, init_optimizer, \11    save_tensor, get_ss_l2, select_op_variables, save_tensor_z12# -------------------------  -------------------------13if not resume:14    copy_tree(cwd+ '/DC', folder+ '/DC')15    with open(folder+'hyperparameters.txt', 'w') as f:16        json.dump(args.__dict__, f, indent=2)17    if not os.path.exists(folder):18        os.makedirs(folder)19    else:20        print('FOLDER ALREADY EXISTS')21# -------------------------  -------------------------22w = 123w1 = 124w2 = 0.525w3 = 0.526w_o = 027tau =1.528remove_things = False29num_seeds = 130def main():31    global soft, w, w1, w2, w3, w_o, init_iter, base_resize, expand_size, n_soft_elements, tau, remove_things, num_seeds32    if args.complex_background:33        base_pattern, background, background_img = load_base_pattern(pattern_filename, args.tiled, base_resize, blur=False, complex_background=True)34        background_img = background_img.cuda()35    else:36        base_pattern, background = load_base_pattern(pattern_filename, args.tiled, base_resize, blur=False)37        background_img = None38    elements = load_element(element_filename, size=base_resize)39    parameters = init_element_pos(number)40    constants = render_constants(number[0]*number[1], scale_x_single, scale_y_single, base_resize, n_soft_elements, expand_size, color=args.color)41    base_pattern, background, mean_c, std_c = make_cuda([base_pattern, background, mean, std])42    base_pattern = vgg_normalize(base_pattern, mean_c[:,:3], std_c[:,:3])43    c_variables = make_cuda(parameters+constants)44    init_c_variables = []45    for i in c_variables:46        init_c_variables.append(i.clone().detach())47    op_variables = select_op_variables(n_elements, args, c_variables)48    if args.tiled:49        base_resize = [512, 512]50    background.requires_grad = True51    if non_white_background:52        op_variables = op_variables+[background, ]53    optimizer, scheduler = init_optimizer(op_variables, args.lr)54    # ----------------- remove unnecessary ------------------------------55    def remove_unecessary(c_variables):56        if not args.color:57            c_variables = remove_invisible_z(c_variables, 2000)58        c_variables = sort_by_z(c_variables)59        element_classes = torch.argmax(c_variables[7], dim=1)60        distance_matrix = d2_distance_matrix(torch.cat([c_variables[0][:, None]/10, c_variables[1][:, None]/10], dim=1))61        idx = select_distance_matrix_idx(distance_matrix, element_classes, c_variables[5], [10/256, 20/256, 10/256, 10/256, 20/256, 15/256, 15/256, 30/256, 30/256], min_dist=10/256)62        new_c_variables = []63        for i in c_variables:64            new_c_variables.append(i[idx])65        c_variables = make_cuda(new_c_variables)66        op_variables = select_op_variables(n_elements, args, c_variables)67        if non_white_background:68            op_variables = op_variables + [background, ]69        optimizer, scheduler = init_optimizer(op_variables, args.lr*0.1)70        return c_variables, optimizer, scheduler71    # ----------------- init ------------------------------72    c_variables[6].requires_grad = False73    if resume:74        load_iter = args.resume_int75        PATH = folder + 'checkpoint_' + str(load_iter)76        checkpoint = torch.load(PATH)77        c_variables = checkpoint['c_variables']78        background = checkpoint['background']79        if args.seed:80            c_variables_1 = c_variables.copy()81            c_variables_1[6] = c_variables_1[6] / 282            c_variables_2 = []83            for c in init_c_variables:84                c_variables_2.append(c.clone().detach())85            c_variables = []86            for idx in range(len(c_variables_1)):87                c_variables.append(torch.cat([c_variables_1[idx], c_variables_2[idx], ]))88        c_variables = make_cuda(c_variables)89        torch.cuda.empty_cache()90        op_variables = select_op_variables(n_elements, args, c_variables)91        background.requires_grad = True92        if non_white_background:93            op_variables = op_variables + [background, ]94        optimizer, scheduler = init_optimizer(op_variables, args.lr)95        # optimizer.load_state_dict(checkpoint['optimizer_state_dict'])96        # scheduler = CyclicLR(optimizer, base_lr=lr, max_lr=lr * 4, cycle_momentum=False, mode='exp_range',97        #                      step_size_up=1500)98        # ----------------------declare variables -------------------99        init_iter = load_iter-1100        w = 1101        w1 = 1102        w2 = 1103        w3 = 1.0104        w_o = 0105        # tau = tau + 10106        if args.layers:107            c_variables[6].requires_grad = True108    current_soft = False109    base_pattern = vgg_renormalize(base_pattern, mean_c[:, :3], std_c[:, :3])110    save_tensor(folder, 'BASE', base_pattern)111    ss = get_ss_l2(base_pattern)112    start_time = time.time()113    for i in range(init_iter+1, init_iter + iter):114        # tau = 0.1115        render, overlap = composite_layers(elements, c_variables, background, n_elements, expand_size, current_soft, tau, color=args.color, background_img=background_img)116        ws = [w, w1, w2, w3, w_o]117        loss = loss_fn_l2(render, overlap, ss, ws) + overlap*w_o118        if i%10==0:119            save_tensor(folder, str(i), render)120            print('iteration {} '.format(i), (loss).to('cpu'), (overlap*(w_o)))121        optimizer.zero_grad()122        loss.backward()123        optimizer.step()124        ## Everything below this is to improve convergence125        if (i- 100000)==500 :126            w_o = 0#(10**-7)127            if args.layers:128                c_variables[6].requires_grad = True129        if i%3000==0:130            # tau = tau +1131            print(torch.exp(c_variables[6]/25))132            if args.layers:133                c_variables[6].requires_grad = True134            save_tensor_z(folder, str(i)+'_z', render, c_variables[0]/10, c_variables[1]/10, c_variables[6], size=128)135            save_tensor_z(folder, str(i)+'_idx', render, c_variables[0]/10, c_variables[1]/10, torch.arange(c_variables[6].shape[0]), size=128)136            torch.save({137                'c_variables': c_variables,138                'optimizer_state_dict': optimizer.state_dict(),139                'base_size':base_resize,140                'xy_scale_factor':10,141                'soft':current_soft,142                'background':background}, folder+'checkpoint_'+str(i))143            w1 = min(w1 + 0.05, 1.5)144            w1 = min(w1 + 0.05, 1.5)145            w2 = max(w2 - 0.05, 0.2)146            w3 = max(w3 - 0.05, 0.2)147        if ((i- 100000)%8000 - 4000 == 500 or  (i- 100000)%8000== 500)and remove_things:148            w_o = 0149            for g in optimizer.param_groups:150                g['lr'] = args.lr151        if (i- 100000)%8000==4000 and remove_things:152            w_o = 0153            elapsed_time = time.time() - start_time154            print(time.strftime("%H:%M:%S", time.gmtime(elapsed_time)))155            print(elapsed_time)156            c_variables, optimizer, scheduler = remove_unecessary(c_variables)157            print(torch.nn.functional.sigmoid(c_variables[5]/10))158            render, overlap = composite_layers(elements, c_variables, background, n_elements, expand_size, current_soft,159                                               tau, color=args.color, background_img=background_img)160            save_tensor(folder, 'postremoval_'+str(i), render)161        if args.seed and num_seeds>0:162            if (i- 100000)%8000==0:163                num_seeds = num_seeds - 1164                torch.save({165                    'c_variables': c_variables,166                    'optimizer_state_dict': optimizer.state_dict(),167                    'base_size': base_resize,168                    'xy_scale_factor': 10,169                    'soft': current_soft,170                    'background': background}, folder + 'checkpoint_' + str(i))171                if args.rotation:172                    c_variables_1 = c_variables.copy()173                    c_variables_1[6] = c_variables_1[6]*2/3174                    # c_variables_1[7] = c_variables_1[7]175                    c_variables_2 = c_variables.copy()176                    c_variables_2[2] = c_variables_2[2] + (90/(1.5*3.14159*57.2958))177                    c_variables_2[6] = c_variables_2[6]*2/3178                    # c_variables_2[7] = torch.ones_like(c_variables_2[7])#/4179                    c_variables_3 = c_variables.copy()180                    c_variables_3[2] = c_variables_3[2] + (180/(1.5*3.14159*57.2958))181                    c_variables_3[6] = c_variables_3[6]*2/3182                    # c_variables_3[7] = torch.ones_like(c_variables_3[7])#/4183                    c_variables_4 = c_variables.copy()184                    c_variables_4[2] = c_variables_4[2] + (270/(1.5*3.14159*57.2958))185                    c_variables_4[6] = c_variables_4[6]*2/3186                    c_variables = []187                    for idx in range(len(c_variables_1)):188                        c_variables.append(torch.cat([c_variables_1[idx], c_variables_2[idx], c_variables_3[idx], c_variables_4[idx]]))189                    c_variables = make_cuda(c_variables)190                    op_variables = select_op_variables(n_elements, args, c_variables)191                    render, overlap = composite_layers(elements, c_variables, background, n_elements, expand_size, current_soft,192                                                       tau, color=args.color, background_img=background_img)193                    save_tensor(folder, 'postseeding_'+str(i), render)194                    optimizer, scheduler = init_optimizer(op_variables+[background, ], args.lr*0.1)195                    remove_things = True196                else:197                    c_variables_1 = c_variables.copy()198                    if args.color:199                        c_variables_1[7] = torch.ones_like(c_variables_1[7])200                        c_variables_2 = []201                        for c in init_c_variables:202                            c_variables_2.append(c.clone().detach())203                        c_variables = []204                        for idx in range(len(c_variables_1)):205                            c_variables.append(torch.cat([c_variables_1[idx], c_variables_2[idx],]))206                    else:207                        c_variables_1[6] = c_variables_1[6] / 2208                        c_variables_2 = []209                        for c in init_c_variables:210                            c_variables_2.append(c.clone().detach())211                        c_variables = []212                        for idx in range(len(c_variables_1)):213                            c_variables.append(torch.cat([c_variables_1[idx], c_variables_2[idx],]))214                    c_variables = make_cuda(c_variables)215                    op_variables = select_op_variables(n_elements, args, c_variables)216                    render, overlap = composite_layers(elements, c_variables, background, n_elements, expand_size, current_soft,217                                                       tau, color=args.color)218                    save_tensor(folder, 'postseeding_'+str(i), render)219                    if non_white_background:220                        op_variables = op_variables + [background, ]221                    optimizer, scheduler = init_optimizer(op_variables, args.lr*0.1)222                    remove_things = True223                w_o = 0224if __name__ == '__main__':...

constants.py

Source:constants.py

1import math2import os3from DC.common import make_tensor4from DC.config import args5from DC.scales import all_scales6# matplotlib.use('Agg')7# ------------------------- variables -------------------------8cwd = os.getcwd()9pattern = args.pattern10n_elements = 111n_soft_elements = int(len(all_scales[pattern])/2)12folder_root = args.folder13if folder_root == '':14    folder_root = cwd#os.path.dirname(cwd)15element_filename = []16for i in range(n_soft_elements):17    element_filename.append('{}/data/{}/elements/{}.png'.format(folder_root, pattern, i+1))18pattern_filename = '{}/data/{}/pattern.png'.format(folder_root, pattern)19folder = "{}/logs/expansion/{}_{}/".format(folder_root, pattern, args.version)20non_white_background = args.non_white21# number = [4, 4]22scale_x_single = all_scales[pattern][0::2]23scale_y_single = all_scales[pattern][1::2]24# ------------------------- calculated constants -------------------------25base_resize = [args.base_size, args.base_size]26expand_size = [1, 3, base_resize[0], base_resize[1]]27if args.expand:28    expand_size = [1, 3, base_resize[0]*2, base_resize[0]*2]29number = [int(1/(scale_y_single[0]* base_resize[1]/expand_size[3])), int(1/(scale_x_single[0]* base_resize[0]/expand_size[2]))]30number[0] = args.sample#(int(number[0] / n_elements)+1) * n_elements31number[1] = args.sample#(int(number[1] / n_elements)+1) * n_elements32repeats = int(number[0] * number[1] / n_elements)33print('Number of elements:', number)34expand_size[0] = repeats35# ------------------------- initializations and constants -------------------------36lr = args.lr37iter = args.num_iter38LBFGS = False39multi = True40soft = args.soft41smooth = False42init_iter = 10000043resume = args.resume44d_threshold = (math.sqrt((max(scale_x_single) * base_resize[0]) ** 2 + (max(scale_y_single) * base_resize[1]) ** 2) -20) / 245mean = make_tensor([0.485, 0.456, 0.406, 0])[None,:,None, None]46std = make_tensor([0.229, 0.224, 0.225, 1])[None,:,None, None]47content_layers = ['pool_4', 'relu4_2',]# 'relu5_2']48content_weights = {'pool_4': 1.0, 'relu4_2': 1.0,}# 'relu5_2': 1.0}49style_layers = ['conv_1', 'conv_2', 'conv_3', 'conv_4']#, 'conv_5']50style_weights = {'conv_1': 0.2, 'conv_2': 0.2, 'conv_3': 0.2, 'conv_4': 0.2}#, 'conv_5': 200}51hist_layers = ['relu4_2', 'conv_1']...

hyperopt.py

Source:hyperopt.py

1from pytorch_ML.hyperopt import Hyperopt2import numpy as np3from numpy import random4from filet.mask_discriminator.transformations import PreProcessor_Crop_n_Resize_Box5class Mask_Hyperopt(Hyperopt):6    def __init__(self,base_lr,**kwargs):7        super().__init__(**kwargs)8        self.base_lr = base_lr9    def suggest_hyper_dict(self):10        '''11        Overloaded from base12        '''13        base_resize = (400,620)14        resize_scales = [0.33,0.5,0.75,1]15        chosen_resize_scale_index = np.random.randint(0,len(resize_scales))16        chosen_scale = resize_scales[chosen_resize_scale_index]17        resize_tup = [int(chosen_scale*base_resize[0]),int(chosen_scale * base_resize[1])]18        is_resize_sq = np.random.randint(0, 2)19        if is_resize_sq:20            resize_tup[0] = ( resize_tup[0] + resize_tup[1] ) // 221            resize_tup[1] = resize_tup[0]22        pad_choices = [5*i+25 for i in range(6)]23        pad = pad_choices[np.random.randint(0,len(pad_choices))]24        generated_values = {25            "optimizer" : { "lr": self.base_lr * random.uniform(1,100), "momentum": random.uniform(0.1, 0.6)},26            "scheduler" : {'gamma' : None},27            "loss" : {},28            "net" : {'two_layer_head' : random.random()>0.5,},29            'resize' : resize_tup,30            'pad' : pad,31        }32        lr_half_time = random.uniform(1000, 7500)33        generated_values['scheduler']["gamma"] = np.exp(-np.log(2)/lr_half_time)34        return generated_values35    def construct_trainer_objs(self,hyper):36        '''37        builds model, optimizer,scheduler and other objects needed for training. Must be deterministic given hyper_dict38        '''39        hyper_objs = super().construct_trainer_objs(hyper)40        dt = self.dt41        prep = PreProcessor_Crop_n_Resize_Box(resize_dims=hyper['resize'],pad=hyper['pad'],mean=[0.2010, 0.1944, 0.2488, 0.0000],std=[0.224, 0.224, 0.224, 1])42        dt.dataset.set_preprocessor(prep)43        hyper_objs['dt'] = dt...

Automation Testing Tutorials

Learn to execute automation testing from scratch with LambdaTest Learning Hub. Right from setting up the prerequisites to run your first automation test, to following best practices and diving deeper into advanced test scenarios. LambdaTest Learning Hubs compile a list of step-by-step guides to help you be proficient with different test automation frameworks i.e. Selenium, Cypress, TestNG etc.