initial commit

download_all.py

+#!/usr/bin/env python
+
+from bob.extension.download import download_and_unzip
+import os
+
+# Test if we are in the CI
+base_url = (
+    "https://www.idiap.ch/software/bob/data/bob/"
+    if os.environ.get("CI_PROJECT_ID") is None
+    else "http://www.idiap.ch/public/data/bob/"
+)
+
+
+all_urls = [
+    "bob.learn.tensorflow/master/densenet-161-imagenet.tar.xz",
+    "bob.db.oulunpu/master/oulunpu-mtcnn-annotations.tar.xz",
+    "bob.db.replaymobile/master/replaymobile-mtcnn-annotations.tar.xz",
+    "bob.db.swan/master/swan-mtcnn-annotations.tar.xz",
+    "bob.db.batl/master/WMCA_annotations_50_frames.tar.xz",
+]
+
+fldr = "downloads"
+os.makedirs(fldr, exist_ok=True)
+for path in all_urls:
+    filename = f"{fldr}/{os.path.basename(path)}"
+    url = [f"{base_url}{path}"]
+    print(f"Downloading {url} and saving in {filename}")
+    download_and_unzip(url, filename)
+print("Finished Downloading!")

evaluate.sh

+#!/bin/bash
+
+set -ex
+
+databases=(oulunpu replaymobile swan batl)
+protocols_1=(Protocol_1 grandtest pad_p2_face_f1 grandtest_color_50_PrintReplay)
+protocols_2=(Protocol_1 grandtest pad_p2_face_f1 grandtest-color-50-PrintReplay)
+
+for ((i=0;i<${#databases[@]};i++)); do
+    bob pad metrics --eval --decimal 2 \
+    results/${databases[i]}/${protocols_1[i]}/{deep_pix_bis,deep_pix_bis_pruned_by_replaymobile,deep_pix_bis_pruned_by_swan,deep_pix_bis_pruned_by_batl,deep_pix_bis_pruned_by_ijbc}/${protocols_2[i]}/scores/scores-{dev,eval} \
+    --legends deep_pix_bis,deep_pix_bis_pruned_by_replaymobile,deep_pix_bis_pruned_by_swan,deep_pix_bis_pruned_by_batl,deep_pix_bis_pruned_by_ijbc
+done

requirements.txt

+setuptools
+numpy
+bob.extension
\ No newline at end of file

run_part1.sh

+#!/bin/bash
+
+# The experiments are done in 3 stages:
+# 1. Caching where all the face PAD datasets are cached for faster training and evaluation in future.
+# 2. Training deep_pix_bis on oulunpu
+# 3. Testing deep_pix_bis on all 4 datasets.
+
+set -ex
+
+databases=(oulunpu replaymobile swan batl)
+protocols=(Protocol_1 grandtest pad_p2_face_f1 grandtest_color_50_PrintReplay)
+algorithms=(deep_pix_bis)
+BOB_SCRIPT=$(which bob)
+TRAIN_CACHE_JOBS=()
+PREDICT_CACHE_JOBS=()
+TRAIN_JOBS=()
+
+# jman options for running on SGE
+LOCAL_OPT=""
+SUBMIT_OPTS="--queue gpu"
+
+# # jman options for running locally
+# LOCAL_OPT="--local"
+# SUBMIT_OPTS=""
+
+# Cache the datasets:
+
+TRAIN_CACHE_JOBS+=($(jman ${LOCAL_OPT} submit ${SUBMIT_OPTS} --print-id --stop-on-failure --name "cache_train_data" \
+${BOB_SCRIPT} tf cache -vvv --mode train \
+bob.paper.icassp2020_domain_guided_pruning.{\
+deep_pix_bis,\
+estimator,\
+oulunpu,\
+Protocol_1,\
+train,\
+load_data_with_normalizer,\
+input_fn}))
+
+TRAIN_CACHE_JOBS+=($(jman ${LOCAL_OPT} submit ${SUBMIT_OPTS} --print-id --stop-on-failure --name "cache_eval_data" \
+${BOB_SCRIPT} tf cache -vvv --mode eval \
+bob.paper.icassp2020_domain_guided_pruning.{\
+deep_pix_bis,\
+estimator,\
+oulunpu,\
+Protocol_1,\
+dev,\
+load_data_with_normalizer,\
+input_fn}))
+
+for ((i=0;i<${#databases[@]};i++)); do
+    PREDICT_CACHE_JOBS+=($(jman ${LOCAL_OPT} submit ${SUBMIT_OPTS} --print-id --stop-on-failure --name "cache_${databases[i]}" \
+${BOB_SCRIPT} tf cache -vvv --mode infer \
+bob.paper.icassp2020_domain_guided_pruning.{\
+deep_pix_bis,\
+estimator,\
+${databases[i]},\
+${protocols[i]},\
+train_dev_eval,\
+load_data_without_normalizer,\
+input_fn}))
+
+done
+
+
+# train the models
+
+for algorithm in ${algorithms[@]}; do
+    TRAIN_JOBS+=($(jman ${LOCAL_OPT} submit ${SUBMIT_OPTS} --print-id --stop-on-failure --name "train_${algorithm}" \
+--dependencies ${TRAIN_CACHE_JOBS[@]} -- \
+${BOB_SCRIPT} tf train -vvv \
+bob.paper.icassp2020_domain_guided_pruning.{\
+${algorithm},\
+estimator,\
+oulunpu,\
+Protocol_1,\
+train,\
+load_data_with_normalizer,\
+input_fn}))
+
+    TRAIN_JOBS+=($(jman ${LOCAL_OPT} submit ${SUBMIT_OPTS} --print-id --stop-on-failure --name "eval_${algorithm}" \
+--dependencies ${TRAIN_CACHE_JOBS[@]} -- \
+${BOB_SCRIPT} tf eval -vvv \
+bob.paper.icassp2020_domain_guided_pruning.{\
+${algorithm},\
+estimator,\
+oulunpu,\
+Protocol_1,\
+dev,\
+load_data_with_normalizer,\
+input_fn}))
+
+done
+
+# run the prediction
+
+for algorithm in ${algorithms[@]}; do
+for ((i=0;i<${#databases[@]};i++)); do
+
+    SUB_DIR="$(pwd)/results/${databases[i]}/${protocols[i]}/${algorithm}"
+
+    LAST_JOB=$(jman ${LOCAL_OPT} submit ${SUBMIT_OPTS} --print-id --stop-on-failure --name "predict_${algorithm}_${databases[i]}" \
+--dependencies ${PREDICT_CACHE_JOBS[@]} ${TRAIN_JOBS[@]} -- \
+${BOB_SCRIPT} tf predict -vvv \
+bob.paper.icassp2020_domain_guided_pruning.{\
+${algorithm},\
+estimator,\
+${databases[i]},\
+${protocols[i]},\
+train_dev_eval,\
+load_data_without_normalizer,\
+input_fn} \
+--checkpoint-path results/${algorithm}/eval \
+--output-dir ${SUB_DIR}/predictions)
+
+    jman ${LOCAL_OPT} submit ${SUBMIT_OPTS} --print-id --stop-on-failure --name "score_${algorithm}_${databases[i]}" \
+--dependencies ${LAST_JOB} -- \
+$(which spoof.py) -vvv \
+bob.paper.icassp2020_domain_guided_pruning.{\
+${algorithm},\
+estimator,\
+${databases[i]},\
+${protocols[i]},\
+train_dev_eval,\
+load_data_with_normalizer,\
+input_fn,\
+pad_video_predictions} \
+--sub-directory ${SUB_DIR} \
+--force
+
+done
+done

version.txt

+0.0.1b0
\ No newline at end of file