def full_energy_batch(out_dir, out_filename, batch_split, gmm1=None, query_tup=None, include_neg=True, use_scaling=True):
import dog_walking_gmm as dwg
import irsg_querygen as qgen
if gmm1 is None:
gmm1 = get_dw_gmms()
if query_tup is None:
query_str = 'dog_walker holding leash attached_to dog'
query_obj = qgen.gen_two_rel_chain(query_str)
query_tup = (query_obj, query_str)
query = qgen.gen_two_rel_chain('dog_walker holding leash attached_to dog')
batches = []
if batch_split == 'pos':
print(' Running PSU images')
pos_test = energy_batch('pos_test', query_tup[0], gmm1, verbose=True, use_scaling=use_scaling)
batches.append(('pos', 'True', 'True', pos_test))
elif batch_split == 'stanford':
print(' Running Stanford images')
res = energy_batch(batch_split, query_tup[0], gmm1, verbose=True, use_scaling=use_scaling)
batches.append(('pos', 'True', 'True', res))
else:
print('unknown batch split type "{}"'.format(batch_split))
if batch_split == 'neg' or include_neg:
print(' Running negative images (No person, No dog)')
neg_np_nd = energy_batch('neg_np_nd', query_tup[0], gmm1, verbose=True, use_scaling=use_scaling)
batches.append(('neg', 'False', 'False', neg_np_nd))
print(' Running negative images (No person, dog)')
neg_np_pd = energy_batch('neg_np_pd', query_tup[0], gmm1, verbose=True, use_scaling=use_scaling)
batches.append(('neg', 'False', 'True', neg_np_pd))
print(' Running negative images (person, No dog)')
neg_pp_nd = energy_batch('neg_pp_nd', query_tup[0], gmm1, verbose=True, use_scaling=use_scaling)
batches.append(('neg', 'True', 'False', neg_pp_nd))
print(' Running negative images (person, dog)')
neg_pp_pd = energy_batch('neg_pp_pd', query_tup[0], gmm1, verbose=True, use_scaling=use_scaling)
batches.append(('neg', 'True', 'True', neg_pp_pd))
# store all the scores
f = open(out_dir + out_filename, 'wb')
for batch in batches:
pos_neg_flag = batch[0]
person_flag = batch[1]
dog_flag = batch[2]
results = batch[3]
prefix = '{}, {}, {}, '.format(pos_neg_flag, person_flag, dog_flag)
for result in results:
line = prefix + '{}, {:4f}\n'.format(result[0], result[1])
f.write(line)
f.close()
def full_viz_plot_batch(out_dir, batch_split, gmms=None, query_tup=None, include_neg=False, use_scaling=True):
import irsg_querygen as qgen
if gmms is None:
gmms = get_dw_gmms()
if query_tup is None:
query_str = 'dog_walker holding leash attached_to dog'
query_obj = qgen.gen_two_rel_chain(query_str)
query_tup = (query_obj, query_str)
if batch_split == 'pos':
print(' Running PSU images')
viz_plot_batch('pos_test', query_tup, gmms, out_dir, verbose=True, use_scaling=use_scaling)
elif batch_split == 'stanford':
print(' Running Stanford images')
viz_plot_batch(batch_split, query_tup, gmms, out_dir, verbose=True, use_scaling=use_scaling)
else:
return
if include_neg:
print(' Running negative images (No person, No dog)')
viz_plot_batch('neg_np_nd', query_tup, gmms, out_dir, verbose=True, use_scaling=use_scaling)
print(' Running negative images (No person, dog)')
viz_plot_batch('neg_np_pd', query_tup, gmms, out_dir, verbose=True, use_scaling=use_scaling)
print(' Running negative images (person, No dog)')
viz_plot_batch('neg_pp_nd', query_tup, gmms, out_dir, verbose=True, use_scaling=use_scaling)
print(' Running negative images (person, dog)')
viz_plot_batch('neg_pp_pd', query_tup, gmms, out_dir, verbose=True, use_scaling=use_scaling)
def full_viz_batch(out_dir, out_filename, batch_split, gmms=None, query_tup=None):
import irsg_querygen as qgen
if gmms is None:
gmms = get_dw_gmms()
if query_tup is None:
query_str = 'dog_walker holding leash attached_to dog'
query_obj = qgen.gen_two_rel_chain(query_str)
query_tup = (query_obj, query_str)
lines = []
if batch_split == 'pos':
print(' Running PSU images')
pos_test = viz_batch('pos_test', query_tup, gmms, verbose=True)
lines.extend(pos_test)
elif batch_split == 'stanford':
print(' Running Stanford images')
res = viz_batch(batch_split, query_tup, gmms, verbose=True)
else:
print(' Running negative images (No person, No dog)')
neg_np_nd = viz_batch('neg_np_nd', query_tup, gmms, verbose=True)
lines.extend(neg_np_nd)
print(' Running negative images (No person, dog)')
neg_np_pd = viz_batch('neg_np_pd', query_tup, gmms, verbose=True)
lines.extend(neg_np_pd)
print(' Running negative images (person, No dog)')
neg_pp_nd = viz_batch('neg_pp_nd', query_tup, gmms, verbose=True)
lines.extend(neg_pp_nd)
print(' Running negative images (person, dog)')
neg_pp_pd = viz_batch('neg_pp_pd', query_tup, gmms, verbose=True)
lines.extend(neg_pp_pd)
# store all the viz data
fq_filename = out_dir
if fq_filename[-1] != '/':
fq_filename += '/'
fq_filename += out_filename
f = open(fq_filename, 'wb')
for line in lines:
f.write(line)
f.close()
def viz_plot(outdir='/home/econser/School/research/output/', filename='dog-walking301.csv', image_type='pos_test', gmms=None, query_tup=None, use_scaling=True):
import opengm as ogm
import plot_utils as pt; reload(pt)
import irsg_querygen as qgen
if gmms is None:
gmms = get_dw_gmms()
image_num = filename_to_id(filename)
if query_tup is None:
query_str = 'dog_walker holding leash attached_to dog'
query_obj = qgen.gen_two_rel_chain(query_str)
query_tup = (query_obj, query_str)
pregen_dir = dir_map[image_type]
classes = [obj.names for obj in query_tup[0].objects]
pg_comps = get_pregen_components(filename, pregen_dir, gmms, classes)
pgm, pgm_to_sg = gen_factor_graph(query_tup[0], pg_comps, use_scaling=use_scaling)
energy, best_box_ixs, marginals = do_inference(pgm)
image_subdir = 'psu_dw/DogWalkingSituationNegativeExamples'
if image_type == 'neg_np_nd':
image_subdir += '/NoDogNoPerson'
elif image_type == 'neg_np_pd':
image_subdir += '/DogNoPerson'
elif image_type == 'neg_pp_nd':
image_subdir += '/PersonNoDog'
elif image_type == 'neg_pp_pd':
image_subdir += '/DogAndPerson'
elif image_type == 'pos_test' or image_type == 'pos_train':
image_subdir = 'psu_dw/PortlandSimpleDogWalking'
elif image_type == 'stanford':
image_subdir = 'StanfordSimpleDogWalking'
image_filename = id_to_filename(image_num, image_type)
pt.draw_best_objects('/home/econser/School/research/images/'+image_subdir, pg_comps, best_box_ixs, energy, out_dir=outdir, out_filename='viz_'+image_filename.split('.')[0]+'.png')
def viz_csv(filename='dog-walking301.csv', image_type='pos_test', gmms=None, query_tup=None):
import irsg_querygen as qgen
import opengm as ogm
if gmms is None:
gmms = get_dw_gmms()
image_num = filename_to_id(filename)
if query_tup is None:
query_str = 'dog_walker holding leash attached_to dog'
query_obj = qgen.gen_two_rel_chain(query_str)
query_tup = (query_obj, query_str)
pregen_dir = dir_map[image_type]
classes = [obj.names for obj in query_tup[0].objects]
pg_comps = get_pregen_components(filename, pregen_dir, gmms, classes)
pgm, pgm_to_sg = gen_factor_graph(query_tup[0], pg_comps)
energy, best_box_ixs, marginals = do_inference(pgm)
# generate the csv data
# 0 : image ID
viz_data = '{:03d}, 0, "{}"\n'.format(image_num, query_tup[1])
# 1 : object ix, obj name, bbox coords
for pgm_obj_ix, bbox_ix in enumerate(best_box_ixs):
sg_obj_ix = pgm_to_sg[pgm_obj_ix]
sg_obj = pg_comps.unary_components[sg_obj_ix]
obj_name = sg_obj.name
b = sg_obj.boxes[bbox_ix]
viz_data += '{:03d}, 1, {}, "{}", {}, {}, {}, {}\n'.format(image_num, sg_obj_ix, obj_name, int(b[0]), int(b[1]), int(b[2]), int(b[3]))
# 2 : energy
viz_data += '{:03d}, 2, {:0.4f}\n'.format(image_num, energy)
# 3 : match/no match
# TODO: paramaterize the TP IDs
if image_num <= 400:
viz_data += '{:03d}, 3, match\n'.format(image_num)
else:
viz_data += '{:03d}, 3, no match\n'.format(image_num)
# 4 : image subdirectory
image_subdir = 'DogWalkingSituationNegativeExamples'
if image_type == 'neg_np_nd':
image_subdir += '/NoDogNoPerson'
elif image_type == 'neg_np_pd':
image_subdir += '/DogNoPerson'
elif image_type == 'neg_pp_nd':
image_subdir += '/PersonNoDog'
elif image_type == 'neg_pp_pd':
image_subdir += '/DogAndPerson'
elif image_type == 'pos_test' or image_type == 'pos_train':
image_subdir = 'PortlandSimpleDogWalking'
viz_data += '{:03d}, 4, {}\n'.format(image_num, image_subdir)
# 5 : image filename
image_filename = id_to_filename(image_num, image_type)
viz_data += '{:03d}, 5, {}\n'.format(image_num, image_filename)
return viz_data