Defect detection network -- hybrid supervision (kolektor defect data set reproduction)

requirement ：
Python 3.8
PyTorch 1.6, 1.8
CUDA 10.0, 10.1
Data sets ：
KolektorSDD ,[ Download this ]
stay colab Running examples on ：
First the git Clone the source code

!git clone https://github.com/vicoslab/mixed-segdec-net-comind2021

Then download KSDD Data sets

% cd /content/mixed-segdec-net-comind2021
% cd /content/mixed-segdec-net-comind2021/datasets/KSDD
!python download_and_extract.py

Then train / Evaluation model ：
train_net.py It is the main entrance of training and evaluation
models.py Is a network model file
data/dataset_catalog.py It contains the currently supported data sets

python -u train_net.py  \
    --GPU=0 \
    --DATASET=KSDD \
    --RUN_NAME=RUN_NAME \
    --DATASET_PATH=/path/to/dataset \
    --RESULTS_PATH=/path/to/save/results \
    --SAVE_IMAGES=True \
    --DILATE=7 \
    --EPOCHS=50 \
    --LEARNING_RATE=1.0 \
    --DELTA_CLS_LOSS=0.01 \
    --BATCH_SIZE=1 \
    --WEIGHTED_SEG_LOSS=True \
    --WEIGHTED_SEG_LOSS_P=2 \
    --WEIGHTED_SEG_LOSS_MAX=1 \
    --DYN_BALANCED_LOSS=True \
    --GRADIENT_ADJUSTMENT=True \
    --FREQUENCY_SAMPLING=True \
    --TRAIN_NUM=33 \
    --NUM_SEGMENTED=33 \
    --FOLD=0

The details are shown in the figure below ：

（ Set your saved path and training parameters ）
After training , The results are stored in results in ,KSDD The evaluation results of all three aspects need to be combined , Use join_folds_results.py：
python -u join_folds_results.py
–RUN_NAME=SAMPLE_RUN
–RESULTS_PATH=/path/to/save/results
–DATASET=KSDD
（ Pay attention to the path you save to and the path where the dataset is located ）

The above figure is the evaluation result , It can also be used. read_results.py Download the data set and generate the result table of all runs .