From da9757edd9f3ea6b5657152ccf261b4840f8fb83 Mon Sep 17 00:00:00 2001
From: Ketan Kotwal <ketan.kotwal@idiap.ch>
Date: Wed, 7 Oct 2020 13:51:16 +0200
Subject: [PATCH] local file for lightcnn; configs for dataset and expt
---
.../config/patch_pooling_lr.py | 4 +-
bob/paper/nir_patch_pooling/config/run.sh | 2 +-
.../nir_patch_pooling/extractor/__init__.py | 2 +-
.../nir_patch_pooling/extractor/lightcnn9.py | 78 +++++++++++++++++++
.../extractor/patch_pooling_cnn.py | 2 +-
.../script/annotate_database.py | 31 ++++----
.../script/convert_mlfp_database.py | 26 +++----
buildout.cfg | 3 +-
setup.py | 7 +-
9 files changed, 117 insertions(+), 38 deletions(-)
create mode 100644 bob/paper/nir_patch_pooling/extractor/lightcnn9.py
diff --git a/bob/paper/nir_patch_pooling/config/patch_pooling_lr.py b/bob/paper/nir_patch_pooling/config/patch_pooling_lr.py
index eba3995..d8f89b8 100644
--- a/bob/paper/nir_patch_pooling/config/patch_pooling_lr.py
+++ b/bob/paper/nir_patch_pooling/config/patch_pooling_lr.py
@@ -22,15 +22,13 @@ FACE_SIZE = 128
RGB_OUTPUT_FLAG = False
USE_FACE_ALIGNMENT = True
ALIGNMENT_TYPE = "lightcnn"
-MIN_FACE_SIZE = 50
NORMALIZATION_FUNCTION = _norm_func
-NORMALIZATION_FUNCTION_KWARGS = {'n_sigma':3.0, 'norm_method':'MAD'}
+NORMALIZATION_FUNCTION_KWARGS = {'n_sigma':4.0, 'norm_method':'MAD'}
_image_preprocessor = FaceCropAlign(face_size=FACE_SIZE,
rgb_output_flag=RGB_OUTPUT_FLAG,
use_face_alignment=USE_FACE_ALIGNMENT,
alignment_type=ALIGNMENT_TYPE,
- min_face_size=MIN_FACE_SIZE,
normalization_function = NORMALIZATION_FUNCTION,
normalization_function_kwargs = NORMALIZATION_FUNCTION_KWARGS)
diff --git a/bob/paper/nir_patch_pooling/config/run.sh b/bob/paper/nir_patch_pooling/config/run.sh
index 38606a4..9659f6c 100755
--- a/bob/paper/nir_patch_pooling/config/run.sh
+++ b/bob/paper/nir_patch_pooling/config/run.sh
@@ -7,7 +7,7 @@ echo "-----------------------------------"
expt="wmca"
-base_directory="/idiap/temp/kkotwal/icip20/expt"
+base_directory="/idiap/temp/kkotwal/test_icip/expt"
# expt: WMCA
if [[ "${expt}" == "wmca" ]]; then
diff --git a/bob/paper/nir_patch_pooling/extractor/__init__.py b/bob/paper/nir_patch_pooling/extractor/__init__.py
index bb19ad5..c4fb95e 100644
--- a/bob/paper/nir_patch_pooling/extractor/__init__.py
+++ b/bob/paper/nir_patch_pooling/extractor/__init__.py
@@ -1,2 +1,2 @@
+from .lightcnn9 import LightCNN9
from .patch_pooling_cnn import PatchPoolingCNN
-
diff --git a/bob/paper/nir_patch_pooling/extractor/lightcnn9.py b/bob/paper/nir_patch_pooling/extractor/lightcnn9.py
new file mode 100644
index 0000000..4038727
--- /dev/null
+++ b/bob/paper/nir_patch_pooling/extractor/lightcnn9.py
@@ -0,0 +1,78 @@
+'''
+ implement Light CNN
+ @author: Alfred Xiang Wu
+ @date: 2017.07.04
+'''
+
+import math
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+#------------------------------------------------------------------------------
+
+class mfm(nn.Module):
+ def __init__(self, in_channels, out_channels, kernel_size=3, stride=1, padding=1, type=1):
+ super(mfm, self).__init__()
+ self.out_channels = out_channels
+ if type == 1:
+ self.filter = nn.Conv2d(in_channels, 2*out_channels, kernel_size=kernel_size, stride=stride, padding=padding)
+ else:
+ self.filter = nn.Linear(in_channels, 2*out_channels)
+
+ def forward(self, x):
+ x = self.filter(x)
+ out = torch.split(x, self.out_channels, 1)
+ return torch.max(out[0], out[1])
+
+#------------------------------------------------------------------------------
+
+class group(nn.Module):
+ def __init__(self, in_channels, out_channels, kernel_size, stride, padding):
+ super(group, self).__init__()
+ self.conv_a = mfm(in_channels, in_channels, 1, 1, 0)
+ self.conv = mfm(in_channels, out_channels, kernel_size, stride, padding)
+
+ def forward(self, x):
+ x = self.conv_a(x)
+ x = self.conv(x)
+ return x
+
+#------------------------------------------------------------------------------
+
+class LightCNN9(nn.Module):
+
+ def __init__(self, num_classes=79077):
+
+ super(LightCNN9, self).__init__()
+ self.features = nn.Sequential(
+ mfm(1, 48, 5, 1, 2),
+ nn.MaxPool2d(kernel_size=2, stride=2, ceil_mode=True),
+ group(48, 96, 3, 1, 1),
+ nn.MaxPool2d(kernel_size=2, stride=2, ceil_mode=True),
+ group(96, 192, 3, 1, 1),
+ nn.MaxPool2d(kernel_size=2, stride=2, ceil_mode=True),
+ group(192, 128, 3, 1, 1),
+ group(128, 128, 3, 1, 1),
+ nn.MaxPool2d(kernel_size=2, stride=2, ceil_mode=True),
+ )
+ self.fc1 = mfm(8*8*128, 256, type=0)
+ self.fc2 = nn.Linear(256, num_classes)
+ self.sigmoid = nn.Sigmoid()
+
+ def forward(self, x):
+
+ x = self.features(x)
+ x = x.view(x.size(0), -1)
+
+ x = self.fc1(x)
+ x = F.dropout(x, p=0.5, training=self.training)
+ scores = self.fc2(x)
+ scores = self.sigmoid(scores)
+
+ # scores, features
+ return scores, x
+
+#------------------------------------------------------------------------------
+
+
diff --git a/bob/paper/nir_patch_pooling/extractor/patch_pooling_cnn.py b/bob/paper/nir_patch_pooling/extractor/patch_pooling_cnn.py
index 0cc37dd..33e807d 100644
--- a/bob/paper/nir_patch_pooling/extractor/patch_pooling_cnn.py
+++ b/bob/paper/nir_patch_pooling/extractor/patch_pooling_cnn.py
@@ -12,7 +12,7 @@ from PIL import Image
import torch.nn.functional as F
from torch.autograd import Variable
import torchvision.transforms as transforms
-from bob.learn.pytorch.architectures import LightCNN9
+from bob.paper.nir_patch_pooling.extractor import LightCNN9
from bob.bio.base.extractor import Extractor
import logging
diff --git a/bob/paper/nir_patch_pooling/script/annotate_database.py b/bob/paper/nir_patch_pooling/script/annotate_database.py
index 5ab7800..013d3be 100755
--- a/bob/paper/nir_patch_pooling/script/annotate_database.py
+++ b/bob/paper/nir_patch_pooling/script/annotate_database.py
@@ -12,7 +12,7 @@ from bob.io.base import create_directories_safe
import numpy as np
import json
import os, sys
-
+from bob.paper.nir_patch_pooling.config.wmca_mask import database as d1
#------------------------------------------------------------------------------
class Annotator:
@@ -32,18 +32,18 @@ class Annotator:
# find annotations using MTCNN
frame_annotations = detect_face_landmarks_in_image(image, method="mtcnn")
if frame_annotations:
- logger.debug(" --> Found using MTCNN")
+ #print(" --> Found using MTCNN")
return frame_annotations
# else, find annotations using dlib
frame_annotations = detect_face_landmarks_in_image(image, method="dlib")
if frame_annotations:
- logger.debug(" --> Found using dlib")
+ print(" --> Found using dlib")
return frame_annotations
# else, return empty dictionary with warning
- logger.warning(" --> Could not find annotations")
- return {}
+ print(" --> Could not find annotations")
+ return None
#------------------------------------------------------------------------------
@@ -85,21 +85,22 @@ class Annotator:
frame_annotations = self.find_annotations(image)
except Exception as e:
- logger.error("\tException: {}".format(e))
+ print("\tException: {}".format(e))
# copy annotations of previous frame
if (prev_valid_index is not None):
frame_annotations = annotations[prev_valid_index]
- logger.warning("\tCopying annotations of valid previous frame")
+ print("\tCopying annotations of valid previous frame")
else:
- frame_annotations = {}
- logger.warning("\tSetting empty annotations")
+ frame_annotations = None
+ print("\tSetting empty annotations")
- annotations[str(index)] = frame_annotations
- prev_valid_index = str(index)
+ if frame_annotations is not None:
+ annotations[str(index)] = frame_annotations
+ prev_valid_index = str(index)
return annotations
@@ -117,7 +118,7 @@ class AnnotationGenerator:
if not os.path.exists(self.annotation_directory):
os.makedirs(self.annotation_directory)
- logger.warning("Annotation directory created at: {}"\
+ print("Annotation directory created at: {}"\
.format(self.annotation_directory))
self.annotator = Annotator()
@@ -166,14 +167,14 @@ class AnnotationGenerator:
# process each video in the given range
for idx, f in enumerate(self.filelist[start_index:end_index]):
- logger.info("[{:03d}/{:03d}] Sample: {}".format(idx+1, total, f.path))
+ print("[{:03d}/{:03d}] Sample: {}".format(idx+1, total, f.path))
json_filepath = os.path.join(self.annotation_directory, f.path + ".json")
if not os.path.exists(json_filepath):
self.process_video(f)
else:
- logger.info("Annotations exist: {}. Skipping".format(f.path))
+ print("Annotations exist: {}. Skipping".format(f.path))
return
@@ -187,7 +188,7 @@ def main():
.format(__name__))
exit(0)
- database = sys.argv[1]
+ database = d1 # sys.argv[1]
annotation_directory = sys.argv[2]
print("Database: {}. Annotation directory: {}".format(database,\
annotation_directory))
diff --git a/bob/paper/nir_patch_pooling/script/convert_mlfp_database.py b/bob/paper/nir_patch_pooling/script/convert_mlfp_database.py
index c6d4f80..d451f5f 100755
--- a/bob/paper/nir_patch_pooling/script/convert_mlfp_database.py
+++ b/bob/paper/nir_patch_pooling/script/convert_mlfp_database.py
@@ -13,10 +13,6 @@ import numpy as np
from bob.pad.face.preprocessor.FaceCropAlign import detect_face_landmarks_in_image
import json
-import logging
-logger = logging.getLogger(__name__)
-logger.setLevel(logging.INFO)
-
frames_per_video = 20
#------------------------------------------------------------------------------
@@ -99,20 +95,20 @@ class MLFPConvertor:
frame_annotations = self.find_annotations(image)
except Exception as e:
- logger.error("\tException: {}".format(e))
+ print(" --> Exception: {}".format(e))
# copy annotations of previous frame
if (prev_index is not None):
frame_annotations = annotations[prev_index]
- logger.warning("\tCopying annotations of previous frame")
- print("\tCopying annotations of previous frame")
+ print(" --> Copying annotations of previous frame")
else:
- frame_annotations = {}
- logger.warning("\tSetting empty annotations")
+ frame_annotations = None
+ print(" --> No annotations")
- annotations[str(index)] = frame_annotations
- prev_index = str(index)
+ if frame_annotations is not None:
+ annotations[str(index)] = frame_annotations
+ prev_index = str(index)
return annotations
@@ -128,18 +124,18 @@ class MLFPConvertor:
# find annotations using MTCNN
frame_annotations = detect_face_landmarks_in_image(image, method="mtcnn")
if frame_annotations:
- logger.debug(" --> Found using MTCNN")
+ #print(" --> Found using MTCNN")
return frame_annotations
# else, find annotations using dlib
frame_annotations = detect_face_landmarks_in_image(image, method="dlib")
if frame_annotations:
- logger.debug(" --> Found using dlib")
+ #print(" --> Found using dlib")
return frame_annotations
# else, return empty dictionary with warning
- logger.warning(" --> Could not find annotations")
- return {}
+ print(" --> Could not find annotations")
+ return None
#------------------------------------------------------------------------------
diff --git a/buildout.cfg b/buildout.cfg
index 5f27ead..b0fb596 100644
--- a/buildout.cfg
+++ b/buildout.cfg
@@ -13,8 +13,9 @@ verbose = true
develop = .
-eggs = bob.paper.nir_patch_pooling
+eggs = bob.pad.base
bob.db.batl
+ bob.paper.nir_patch_pooling
[sources]
bob.db.batl = git git@gitlab.idiap.ch:bob/bob.db.batl rev=5fecee9158832516c560d000a71ae7272d640506
diff --git a/setup.py b/setup.py
index 28f823c..f2b9b4f 100644
--- a/setup.py
+++ b/setup.py
@@ -34,7 +34,12 @@ setup(
install_requires=install_requires,
entry_points={
- # mlfp, wmca-mask, expt-patch-pooling
+ 'bob.bio.config': [
+ 'wmca-mask = bob.paper.nir_patch_pooling.config.wmca_mask',
+ 'mlfp = bob.paper.nir_patch_pooling.config.mlfp',
+ 'patch-pooling-lr = bob.paper.nir_patch_pooling.config.patch_pooling_lr',
+ ],
+
},
# check classifiers, add and remove as you see fit
--
GitLab