文件列表:

.vscode (0, 2017-08-01)
.vscode\settings.json (32, 2017-08-01)
IQA_BIECON_release (0, 2017-08-01)
IQA_BIECON_release\__init__.py (109, 2017-08-01)
IQA_BIECON_release\config_parser.py (7440, 2017-08-01)
IQA_BIECON_release\configs (0, 2017-08-01)
IQA_BIECON_release\configs\NR_biecon.yaml (1102, 2017-08-01)
IQA_BIECON_release\data_load (0, 2017-08-01)
IQA_BIECON_release\data_load\LIVE.py (2166, 2017-08-01)
IQA_BIECON_release\data_load\TID2008.py (1761, 2017-08-01)
IQA_BIECON_release\data_load\TID2013.py (1761, 2017-08-01)
IQA_BIECON_release\data_load\__init__.py (109, 2017-08-01)
IQA_BIECON_release\data_load\data_loader_IQA.py (39562, 2017-08-01)
IQA_BIECON_release\data_load\dataset.py (14029, 2017-08-01)
IQA_BIECON_release\default_config.yaml (1710, 2017-08-01)
IQA_BIECON_release\layers (0, 2017-08-01)
IQA_BIECON_release\layers\__init__.py (109, 2017-08-01)
IQA_BIECON_release\layers\layers.py (16828, 2017-08-01)
IQA_BIECON_release\layers\normalization.py (9687, 2017-08-01)
IQA_BIECON_release\models (0, 2017-08-01)
IQA_BIECON_release\models\BIECON_base.py (11276, 2017-08-01)
IQA_BIECON_release\models\__init__.py (109, 2017-08-01)
IQA_BIECON_release\models\model_basis.py (12088, 2017-08-01)
IQA_BIECON_release\models\model_record.py (4072, 2017-08-01)
IQA_BIECON_release\optimizer.py (11117, 2017-08-01)
IQA_BIECON_release\ssim.py (2173, 2017-08-01)
IQA_BIECON_release\train_iqa.py (9836, 2017-08-01)
IQA_BIECON_release\trainer.py (27661, 2017-08-01)
IQA_BIECON_release\utils.py (17218, 2017-08-01)
LICENSE (1067, 2017-08-01)
LIVE_IQA.txt (58500, 2017-08-01)
TID2008.txt (117320, 2017-08-01)
TID2013.txt (207550, 2017-08-01)
example.py (374, 2017-08-01)
gen_local_metric_scores.m (1965, 2017-08-01)
metrics (0, 2017-08-01)
metrics\ssim_index.m (5977, 2017-08-01)
... ...

# BIECON A blind image evaluator based on a convolutional neural network (BIECON) is a no-reference image quality assessment method using a CNN. This code implements the system described in the following paper: > J. Kim and S. Lee, “Fully deep blind image quality predictor,” IEEE Journal of Selected Topics in Signal Processing, vol. 11, no. 1, pp. 206–220, Feb. 2017. ## Prerequisites This code was developed and tested with Theano 0.9, CUDA 8.0, and Windows. ## Generating local quality score maps 1. Set `BASE_PATH` to the actual root path of each database. Set `FR_MET_BASEPATH` and `FR_MET_SUBPATH` in `gen_local_metric_scores.m`. 2. For each database, data will be stored in "`FR_MET_BASEPATH` + `FR_MET_SUBPATH`". 3. Run `gen_local_metric_scores.m` using Matlab. We provide a SSIM metric as default. ## Environment setting ### Setting database path: For each database, set `BASE_PATH` to the actual root path of each database in the following files: `IQA_BIECON_release/data_load/LIVE.py`, `IQA_BIECON_release/data_load/TID2008.py`, and `IQA_BIECON_release/data_load/TID2013.py`. (These `BASE_PATH` should be same to the `BASE_PATH` in `gen_local_metric_scores.m`.) ### Setting local quality score map path: Set `FR_MET_BASEPATH` and `FR_MET_SUBPATH_{DB name}` in `IQA_BIECON_release/data_load/data_loader_IQA.py`. {*DB name*} can be *LIVE*, *TID2008*, or *TID2013* (These should be same to those in `gen_local_metric_scores.m`.) Detailed configuration of local quality score maps is set in `NR_biecon.yaml`. - `fr_met`: This describes the name of the full-reference image quality assessment metric. The corresponding local quality score maps must be generated first. ex) SSIM, FSIM ... - `fr_met_scale`: This indicates the scale ratio of the local quality score maps to their original images. This should be same to `image_resize` in `gen_local_metric_scores.m`. - `fr_met_avg`: If True, for each divided image patch, local quality score maps are averaged to be scalar values. Otherwise, patch of local quality score maps are used. This is used to reduce the usage of memory. ## Training BIECON We provide the demo code for training a BIECON model. ```bash python example.py ``` - `tr_te_file`: Store the randomly divided (training and testing) reference image indices in this file. - `snap_path`: This indicates the path to store snapshot files ## Quantitative results BIECON was tested on the full-sets of LIVE IQA, TID2013, and CSIQ databases. During the experiment, we randomly divided the reference images into two subsets, 80% for training and 20% for testing. The correlation coefficients were averaged after the procedure was repeated 10 times while dividing the training and testing sets randomly. |Database |SRCC |PLCC | |---------|:-----:|:-----:| |LIVE |0.9603 | 0.9622| |TID2013 |0.7205 | 0.7650| |CSIQ |0.8251 | 0.8380|

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：