def predict():
if request.method == 'GET':
try:
url = request.args.get('q')
app.logger.debug('url provided - %s', url)
input_tensor = transform_image(read_file(url=url))
values, indices = get_topk(input_tensor)
results = render_prediction(values, indices)
return jsonify(results=results)
except:
app.logger.debug("Error: %s", traceback.print_exc())
return jsonify("invalid image url")
elif request.method == 'POST':
try:
file = request.files['file']
app.logger.debug('file uploaded - %s', file)
url = request.form.get("url", None)
app.logger.debug('url provided - %s', url)
input_tensor = transform_image(read_file(upload=file, url=url))
values, indices = get_topk(input_tensor)
results = render_prediction(values, indices)
return jsonify(results=results)
except:
app.logger.debug("Error: %s", traceback.print_exc())
return jsonify("invalid image")
else:
app.logger.debug("Error: %s", traceback.print_exc())
return jsonify('invalid request')
def run_across_movie(data_dir, subset, algorithm, temporal_link, gpu_id):
affinity_dir = osp.join(data_dir, 'affinity', subset, 'across')
list_file = osp.join(data_dir, 'meta', 'across_{}.json'.format(subset))
pids, gt_list, gt_dict = read_across_movie_meta(list_file)
# read affinity matrix
if temporal_link:
link_type = 'max'
else:
link_type = 'mean'
ct_affmat = read_affmat_across_movies(affinity_dir,
region='face',
data_type='ct',
link_type=link_type)
tt_affmat = read_affmat_across_movies(affinity_dir,
region='body',
data_type='tt',
link_type=link_type)
# run algorithm
if algorithm == 'ppcc':
result = run_ccpp(ct_affmat, tt_affmat, gpu_id)
elif algorithm == 'lp':
result = run_lp(ct_affmat, tt_affmat, gpu_id)
else:
raise ValueError('No such algrothm: {}'.format(algorithm))
# parse results and get performance
ret_dict = affmat2retdict(result, pids)
ret_list = affmat2retlist(result, pids)
mAP = get_mAP(gt_dict, ret_dict)
topk = get_topk(gt_list, ret_list)
print('mAP: {:.4f}\ttop1: {:.4f}\ttop3: {:.4f}\ttop5: {:.4f}'.format(
mAP, topk[0], topk[2], topk[4]))
def run_across_movie(data_dir, subset, data_type, face_ratio):
affinity_dir = osp.join(data_dir, 'affinity', subset, 'across')
list_file = osp.join(data_dir, 'meta', 'across_{}.json'.format(subset))
pids, gt_list, gt_dict = read_across_movie_meta(list_file)
# read affinity matrix
if data_type == 'face':
affmat = read_affmat_across_movies(affinity_dir,
region='face',
data_type='ct')
elif data_type == 'body':
affmat = read_affmat_across_movies(affinity_dir,
region='body',
data_type='ct')
else:
face_affmat = read_affmat_across_movies(affinity_dir,
region='face',
data_type='ct')
body_affmat = read_affmat_across_movies(affinity_dir,
region='body',
data_type='ct')
if data_type == 'ave_fusion':
affmat = face_ratio * face_affmat + (1 - face_ratio) * body_affmat
else:
affmat = np.maximum(face_affmat, body_affmat)
# parse results and get performance
ret_dict = affmat2retdict(affmat, pids)
ret_list = affmat2retlist(affmat, pids)
mAP = get_mAP(gt_dict, ret_dict)
topk = get_topk(gt_list, ret_list)
print('mAP: {:.4f}\ttop1: {:.4f}\ttop3: {:.4f}\ttop5: {:.4f}'.format(
mAP, topk[0], topk[2], topk[4]))
def run_in_movie(data_dir, subset, algorithm, temporal_link, gpu_id):
affinity_dir = osp.join(data_dir, 'affinity', subset, 'in')
list_file = osp.join(data_dir, 'meta', subset + '.json')
mid_list, meta_info = read_meta(list_file)
average_mAP = 0
search_count = 0
average_top1 = 0
average_top3 = 0
average_top5 = 0
for i, mid in enumerate(mid_list):
# read data
tnum = meta_info[mid]['num_tracklet']
pids = meta_info[mid]['pids']
gt_list, gt_dict = parse_label(meta_info, mid)
# read affinity matrix
if temporal_link:
link_type = 'max'
else:
link_type = 'mean'
ct_affmat = read_affmat_of_one_movie(affinity_dir,
mid,
region='face',
data_type='ct',
link_type=link_type)
tt_affmat = read_affmat_of_one_movie(affinity_dir,
mid,
region='body',
data_type='tt',
link_type=link_type)
# run algorithm
if algorithm == 'ppcc':
result = run_ccpp(ct_affmat, tt_affmat, gpu_id)
elif algorithm == 'lp':
result = run_lp(ct_affmat, tt_affmat, gpu_id)
else:
raise ValueError('No such algrothm: {}'.format(algorithm))
# parse results and get performance
ret_dict = affmat2retdict(result, pids)
ret_list = affmat2retlist(result, pids)
mAP = get_mAP(gt_dict, ret_dict)
topk = get_topk(gt_list, ret_list)
average_mAP += mAP * len(pids)
search_count += len(pids)
max_k = len(topk)
if max_k < 3:
top3 = 1
else:
top3 = topk[2]
if max_k < 5:
top5 = 1
else:
top5 = topk[4]
average_top1 += topk[0]
average_top3 += top3
average_top5 += top5
# get average performance
average_mAP = average_mAP / search_count
average_top1 = average_top1 / len(mid_list)
average_top3 = average_top3 / len(mid_list)
average_top5 = average_top5 / len(mid_list)
print(
'Average mAP: {:.4f}\tAverage top1: {:.4f}\tAverage top3: {:.4f}\tAverage top5: {:.4f}'
.format(average_mAP, average_top1, average_top3, average_top5))
test_mi_f1 = f1_score(y_true, y_pred, average='micro')
test_ma_f1 = f1_score(y_true, y_pred, average='macro')
test_accuracy = correct/total
writer.add_scalar('accuracy_test/accuracy_test', test_accuracy, global_batch_counter_test)
writer.add_scalar('accuracy_test/micro_f1_test', test_mi_f1, global_batch_counter_test)
writer.add_scalar('accuracy_test/macro_f1_test', test_ma_f1, global_batch_counter_test)
print('[%d] epoch, [%.3f] training loss, [%.3f] testing loss, [%.3f] testing accuracy'
%(epoch, loss_train/len(train_loader), loss_test/test_idx, test_accuracy))
print('Saving models and optimizer...')
save_model_optimizer(model, optimizer, epoch, global_batch_counter_train, global_batch_counter_test, dir)
print('Saved!')
if bootstrapping_start_epoch and topk_pre_class*num_labels <= len(test_set) * bootstrapping_max_usage:
print('Bootstrapping...')
extra_train_set = get_topk(logits, topk_pre_class, y_pred, test_set)
train_set = ConcatDataset([raw_train_set, extra_train_set])
train_loader = DataLoader(train_set, sampler=RandomSampler(train_set), batch_size=batch_size_train)
topk_pre_class = int(len(train_set)//num_labels * bootstrapping_increase_coef)
loss_train = 0
model.train()
elif not save_freq_n_epoch:
print('Saving models and optimizer...')
save_model_optimizer(model, optimizer, epoch, global_batch_counter_train, global_batch_counter_test, dir)
print('Saved!')
if DEBUG:
break
def run_in_movie(data_dir, subset, data_type, face_ratio):
affinity_dir = osp.join(data_dir, 'affinity', subset, 'in')
list_file = osp.join(data_dir, 'meta', subset + '.json')
mid_list, meta_info = read_meta(list_file)
average_mAP = 0
search_count = 0
average_top1 = 0
average_top3 = 0
average_top5 = 0
for i, mid in enumerate(mid_list):
# read data
tnum = meta_info[mid]['num_tracklet']
pids = meta_info[mid]['pids']
gt_list, gt_dict = parse_label(meta_info, mid)
# read affinity matrix
if data_type == 'face':
affmat = read_affmat_of_one_movie(affinity_dir,
mid,
region='face',
data_type='ct')
elif data_type == 'body':
affmat = read_affmat_of_one_movie(affinity_dir,
mid,
region='body',
data_type='ct')
else:
face_affmat = read_affmat_of_one_movie(affinity_dir,
mid,
region='face',
data_type='ct')
body_affmat = read_affmat_of_one_movie(affinity_dir,
mid,
region='body',
data_type='ct')
if data_type == 'ave_fusion':
affmat = face_ratio * face_affmat + (1 -
face_ratio) * body_affmat
else:
affmat = np.maximum(face_affmat, body_affmat)
# parse results and get performance
ret_dict = affmat2retdict(affmat, pids)
ret_list = affmat2retlist(affmat, pids)
mAP = get_mAP(gt_dict, ret_dict)
topk = get_topk(gt_list, ret_list)
average_mAP += mAP * len(pids)
search_count += len(pids)
max_k = len(topk)
if max_k < 3:
top3 = 1
else:
top3 = topk[2]
if max_k < 5:
top5 = 1
else:
top5 = topk[4]
average_top1 += topk[0]
average_top3 += top3
average_top5 += top5
# get average performance
average_mAP = average_mAP / search_count
average_top1 = average_top1 / len(mid_list)
average_top3 = average_top3 / len(mid_list)
average_top5 = average_top5 / len(mid_list)
print(
'Average mAP: {:.4f}\tAverage top1: {:.4f}\tAverage top3: {:.4f}\tAverage top5: {:.4f}'
.format(average_mAP, average_top1, average_top3, average_top5))