def train_image_classify_svm(cc, cls, C=1.0, gamma=0.0, force_new=False):
current_fold = cc.d.current_fold
filename = config.get_classifier_svm_name(cls, C, gamma, current_fold)
if not force_new and os.path.exists(filename):
print 'svm for class %s, C=%f, gamma=%f exists already' % (cls, C,
gamma)
return
pyr_feat_size = get_pyr_feat_size(cc.L, cc.dense_codebook.shape[0])
cc.tictocer.tic('overall')
print 'compute classifier(C=%f, gamma=%f) for class %s' % (C, gamma, cls)
pos_images = cc.d.get_pos_samples_for_fold_class(cls)
if pos_images.size == 0:
return
neg_images = cc.d.get_neg_samples_for_fold_class(cls, pos_images.size)
# 1. extract all the pyramids
# ======== POSTIVE IMAGES ===========
print 'compute feature vector for positive images'
num_slices = 3
pos_pyrs = np.zeros(
(len(pos_images),
pyr_feat_size + cc.sparse_codebook.shape[0] * (1 + num_slices)))
for idx, img in enumerate(pos_images):
pos_pyrs[idx, :] = get_feature_vector(cc, img)
# ======== NEGATIVE IMAGES ===========
print 'compute feature vector for negative images'
neg_pyrs = np.zeros(
(len(neg_images),
pyr_feat_size + cc.sparse_codebook.shape[0] * (1 + num_slices)))
for idx, img in enumerate(neg_images):
neg_pyrs[idx, :] = get_feature_vector(cc, img)
# 2. Compute SVM for class
X = np.vstack((pos_pyrs, neg_pyrs))
Y = [1] * pos_pyrs.shape[0] + [-1] * neg_pyrs.shape[0]
if X.shape[0] > 0:
print 'train svm for class %s, C=%f, gamma=%f' % (cls, C, gamma)
cc.tictocer.tic()
clf = train_svm(X, Y, kernel='rbf', gamma=gamma, C=C)
print '\ttook %f seconds' % cc.tictocer.toc(quiet=True)
print 'save as', filename
save_svm(clf, filename)
else:
print 'Don\'t compute SVM, no examples given'
print 'training all classifier SVMs took:', cc.tictocer.toc(
'overall', quiet=True), 'seconds on', comm_rank
def train(self, pos, neg, kernel, C): y = [1]*pos.shape[0] + [-1]*neg.shape[0] x = np.concatenate((pos,neg)) model = train_svm(x, y, kernel, C) self.svm = model print 'model.score(C=%d): %f'%(C, model.score(x,y)) table_t = svm_proba(x, model) y2 = np.array(y) y2 = (y2+1)/2 # switch to 0/1 ap,_,_ = Evaluation.compute_cls_pr(table_t[:,1], y2) print 'ap on train set: %f'%ap filename = config.get_classifier_filename(self, self.cls, self.train_dataset) self.svm = model self.save_svm(model, filename) return model
def train_image_classify_svm(cc, cls, C=1.0, gamma=0.0, force_new=False):
current_fold = cc.d.current_fold
filename = config.get_classifier_svm_name(cls, C, gamma, current_fold)
if not force_new and os.path.exists(filename):
print 'svm for class %s, C=%f, gamma=%f exists already'%(cls,C,gamma)
return
pyr_feat_size = get_pyr_feat_size(cc.L, cc.dense_codebook.shape[0])
cc.tictocer.tic('overall')
print 'compute classifier(C=%f, gamma=%f) for class %s'%(C, gamma, cls)
pos_images = cc.d.get_pos_samples_for_fold_class(cls)
if pos_images.size == 0:
return
neg_images = cc.d.get_neg_samples_for_fold_class(cls, pos_images.size)
# 1. extract all the pyramids
# ======== POSTIVE IMAGES ===========
print 'compute feature vector for positive images'
num_slices = 3
pos_pyrs = np.zeros((len(pos_images),pyr_feat_size + cc.sparse_codebook.shape[0]*(1+num_slices)))
for idx, img in enumerate(pos_images):
pos_pyrs[idx, :] = get_feature_vector(cc, img)
# ======== NEGATIVE IMAGES ===========
print 'compute feature vector for negative images'
neg_pyrs = np.zeros((len(neg_images),pyr_feat_size + cc.sparse_codebook.shape[0]*(1+num_slices)))
for idx, img in enumerate(neg_images):
neg_pyrs[idx, :] = get_feature_vector(cc, img)
# 2. Compute SVM for class
X = np.vstack((pos_pyrs, neg_pyrs))
Y = [1]*pos_pyrs.shape[0] + [-1]*neg_pyrs.shape[0]
if X.shape[0] > 0:
print 'train svm for class %s, C=%f, gamma=%f'%(cls,C,gamma)
cc.tictocer.tic()
clf = train_svm(X, Y, kernel='rbf', gamma=gamma, C=C)
print '\ttook %f seconds'%cc.tictocer.toc(quiet=True)
print 'save as', filename
save_svm(clf, filename)
else:
print 'Don\'t compute SVM, no examples given'
print 'training all classifier SVMs took:', cc.tictocer.toc('overall', quiet=True), 'seconds on', comm_rank