Failed to create a directory: logs/koniq/KonCept512/bsz:16 i:1[384,512,3] l:MSE o:1[1];

[SheydaGhb]

I am running koncept512_train_test_py3.ipynb but changed it to use it in pycharm and I get this error : tensorflow.python.framework.errors_impl.InvalidArgumentError: Failed to create a directory: logs/koniq/KonCept512/bsz:16 i:1[384,512,3] l:MSE o:1[1]; Invalid argument [Op:CreateSummaryFileWriter]

could you please help me? is it related to how I call input data?

This how I changed the code :

Setup paths

drive_mount = "" drive_root = "" # persistent storage for dataset images data_root = "" # library install path

from ku import model_helper as mh from ku import applications as apps from ku import tensor_ops as ops from ku import generic as gen from ku import image_utils as iu

import pandas as pd, numpy as np, os from matplotlib import pyplot as plt from munch import Munch

import os import shutil import subprocess

from google.colab import drive

drive.mount(drive_mount)

Make directories

data_root = "C:/Users/sheyd/PycharmProjects/NRIQA_methods/Koniq512/koniq10k_512x384/"

define the KonCeptmodel

ids = pd.read_csv(data_root + 'metadata/koniq10k_distributions_sets.csv')

from keras.models import Model

Build scoring model

base_model, preprocess_fn = apps.get_model_imagenet(apps.InceptionResNetV2) head = apps.fc_layers(base_model.output, name='fc', fc_sizes = [2048, 1024, 256, 1], dropout_rates = [0.25, 0.25, 0.5, 0], batch_norm = 2)

model = Model(inputs = base_model.input, outputs = head)

Parameters of the generator

pre = lambda im: preprocess_fn( iu.ImageAugmenter(im, remap=False).fliplr().result) gen_params = dict(batch_size = 16, data_path = data_root+'images/512x384/', process_fn = pre, input_shape = (384,512,3), inputs = ['image_name'], outputs = ['MOS'])

Wrapper for the model, helps with training and testing

helper = mh.ModelHelper(model, 'KonCept512', ids, loss='MSE', metrics=["MAE", ops.plcc_tf], monitor_metric = 'val_loss', monitor_mode = 'min', multiproc = True, workers = 5, logs_root = drive_root + 'logs/koniq/', models_root = drive_root + 'models/koniq/', gen_params = gen_params)

do validation in memory

valid_gen = helper.make_generator(ids[ids.set=='validation'], batch_size = len(ids[ids.set=='validation'])) valid_data = valid_gen[0][0]

helper.train(valid_gen=valid_data, lr=1e-4, epochs=40) helper.load_model() helper.train(valid_gen=valid_data, lr=1e-4/5, epochs=20)