{"cells":[{"metadata":{},"cell_type":"markdown","source":"DEEPFAKE DETECTION CHALLENGE\n\nGood bye DFDC, many hours of work and learning for all of the competitors, in this case my team barely pass the 0.69 logLoss threshold\n\nMaking public 2 python scripts (several distinct versions submitted) used to submit our work,\n\nhttps://www.kaggle.com/pedromoya/a-euler-in-kaer-morhen?rvi=1\n\nhttps://www.kaggle.com/pedromoya/b-euler-in-kaer-morhen?rvi=1\n\n(native idiom in comments of scripts is spanish, but code is very easy to understand)"},{"metadata":{},"cell_type":"markdown","source":"**remarks in this scripts:\n\ntime control for each video and for the entire notebook submission, \n\nerror catch and log (deactivated for submit), \nfunction like: clip-bandpass filter, \nnormalization (and standarization), \nrescale image in numpy, \ncustomized kernel convolute, \nchrom map (papers below), \nROIs detecion (face and sub-regions), \nfilters,\netc."},{"metadata":{},"cell_type":"markdown","source":"if no-face is detected my team decided to output 0.65 and no 0.5\n\nthe main problem for our approach (very simple CNN and all only frame-based) was overfitting and failure in teach the correct generalization needed"},{"metadata":{},"cell_type":"markdown","source":"papers readed (and some used) are added as data to this notebook"}],"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat":4,"nbformat_minor":4}