def drawTubeBB((mat_file, path_to_data, out_file, tubes, fills)):
just_mat_file = mat_file[mat_file.rindex('/') + 1:]
category = just_mat_file[:just_mat_file.index('_')]
video_id = just_mat_file[just_mat_file.index('_') +
1:just_mat_file.rindex('_')]
shot_id = just_mat_file[just_mat_file.rindex('_') +
1:just_mat_file.rindex('.')]
res = scipy.io.loadmat(mat_file, squeeze_me=True, struct_as_record=False)
res = res['res']
if not hasattr(res, '__iter__'):
res = [res]
num_frames = res[0].bboxes.shape[0]
num_tubes = len(res)
frame_idx_all = tuple([res_curr.bboxes[:, -1] for res_curr in res])
frame_idx_all = reduce(np.intersect1d, frame_idx_all)
assert frame_idx_all.size > 0
idx_idx = len(frame_idx_all) / 2
frame_id = int(frame_idx_all[idx_idx])
print mat_file, category, video_id, shot_id, frame_id
frame_path = youtube.getFramePath(path_to_data, category, video_id,
shot_id, frame_id)
im = Image.open(frame_path)
draw = ImageDraw.Draw(im)
coords = getTubeCoordsForFrame(res, frame_id)
for idx_tube_no, tube_no in enumerate(tubes):
bbox = coords[tube_no]
# bboxes=res[tube_no].bboxes
# bbox=bboxes[bboxes[:,-1]==frame_id,:4].ravel()
# # print bbox.shape
# bbox=bbox-1;
# bbox=[math.floor(x/0.4) for x in bbox];
if len(im.getbands()) == 1:
draw.line(getLineCoords(bbox), width=1, fill=fills[idx_tube_no][0])
else:
draw.line(getLineCoords(bbox), width=1, fill=fills[idx_tube_no])
im = np.array(im)
misc.imsave(out_file, im)
def main():
out_file_crude = '/disk2/januaryExperiments/tube_scoring/iou_crude_info.p'
out_file_bestOfBoth = '/disk2/januaryExperiments/tube_scoring/iou_bestOfBoth.p'
out_file_bestOfBoth_perShot = '/disk2/januaryExperiments/tube_scoring/iou_bestOfBoth_perShot.p'
out_file_file = '/disk2/januaryExperiments/tube_scoring/iou_crude_info.p'
class_labels_map = [('boat', 2), ('train', 9), ('dog', 6), ('cow', 5),
('aeroplane', 0), ('motorbike', 8), ('horse', 7),
('bird', 1), ('car', 3), ('cat', 4)]
avg_iou_pred, avg_iou_best = getAverageIOUPerClassBOB(
out_file_bestOfBoth, class_labels_map)
avg_iou_pred_perShot, _ = getAverageIOUPerClassBOB(
out_file_bestOfBoth_perShot, class_labels_map)
avg_iou_fanyi = getAverageIOUBestFanyi(out_file_crude, class_labels_map)
dict_vals = {}
label_keys = ['Fanyi', 'Shot', 'Frame', 'Best']
avg_ious = [
avg_iou_fanyi, avg_iou_pred_perShot, avg_iou_pred, avg_iou_best
]
xtick_labels = avg_iou_pred.keys()
print xtick_labels
for k in avg_iou_pred:
for idx in range(len(label_keys)):
if label_keys[idx] in dict_vals:
dict_vals[label_keys[idx]].append(avg_ious[idx][k])
else:
dict_vals[label_keys[idx]] = [avg_ious[idx][k]]
# dict_vals['Shot']=avg_iou_fanyi[k]
# dict_vals['Frame']=avg_iou_fanyi[k]
# dict_vals['Best']=avg_iou_fanyi[k]
print k, avg_iou_fanyi[k], avg_iou_pred_perShot[k], avg_iou_pred[
k], avg_iou_best[k]
out_file = '/disk2/januaryExperiments/tube_scoring/avg_iou_comparison.png'
colors = ['r', 'g', 'b', 'y']
visualize.plotGroupBar(out_file,
dict_vals,
xtick_labels,
label_keys,
colors,
ylabel='Average IOU',
title='Average IOU Comparison',
width=0.25,
ylim=[0.2, 0.9])
return
# path_to_txt='/disk2/youtube/categories/gt_list.txt';
params_dict = {}
params_dict['class_labels_map'] = [('boat', 2), ('train', 9), ('dog', 6),
('cow', 5), ('aeroplane', 0),
('motorbike', 8), ('horse', 7),
('bird', 1), ('car', 3), ('cat', 4)]
params_dict[
'out_file'] = '/disk2/januaryExperiments/tube_scoring/gt_data.p'
params_dict[
'path_to_tube_ranks'] = '/disk2/januaryExperiments/tube_scoring/scores_perShot'
params_dict['path_to_mats'] = '/disk2/res11'
params_dict[
'out_file_crude_info'] = '/disk2/januaryExperiments/tube_scoring/iou_crude_info_perShot.p'
params_dict[
'out_file_bestOfBoth'] = '/disk2/januaryExperiments/tube_scoring/iou_bestOfBoth_perShot.p'
params = createParams('saveIOUInfo')
params = params(**params_dict)
script_saveIOUInfo(params)
return
# path_to_data = '/disk2/youtube/categories'
# saveGTData(path_to_txt,class_labels_map,out_file)
# return
[class_labels, class_idx_all] = zip(*class_labels_map)
[meta_info, coords] = pickle.load(open(out_file, 'rb'))
meta_curr = meta_info[0]
gt_box = coords[0]
meta_curr_str = [str(curr) for curr in meta_curr]
class_label = class_labels[class_idx_all.index(meta_curr[0])]
tube_file = os.path.join(path_to_tube_ranks,
'_'.join(meta_curr_str[:3]) + '.p')
mat_file = os.path.join(
path_to_mats, '_'.join([class_label] + meta_curr_str[1:3]) + '.mat')
res = scipy.io.loadmat(mat_file, squeeze_me=True,
struct_as_record=False)['res']
tube_coords = getTubeCoordsForFrame(res, meta_curr[-1])
ious, tubes_gt = getTubeGTOverlap(gt_box, tube_coords)
tubes_ranked, tubes_scores = pickle.load(open(tube_file, 'rb'))
tubes_ranked = list(tubes_ranked)
gt_idx_best_rank, iou_best_rank, rank_idx_best_gt, iou_best_gt = getBestOfBothRank(
ious, tubes_gt, tubes_ranked)
print gt_idx_best_rank, iou_best_rank, rank_idx_best_gt, iou_best_gt
print tubes_ranked
return
frame_path = youtube.getFramePath(path_to_data, class_label, meta_curr[1],
meta_curr[2], meta_curr[3])
im = Image.open(frame_path)
draw = ImageDraw.Draw(im)
for idx_tube, tube in enumerate(tubes_ranked):
draw.line(getLineCoords(tube_coords[tube]),
width=2,
fill=(255, 255, 0))
draw.line(getLineCoords(tube_coords[tubes_ranked[0]]),
width=2,
fill=(255, 0, 0))
# for gt_box in gt_coords:
# print gt_box
draw.line(getLineCoords(gt_box), width=2, fill=(0, 255, 0))
out_file = '/disk2/temp/temp.png'
misc.imsave(out_file, np.array(im))
# gt_boxes=coords[0];
# break
# video_id=;
# shot_id=;
# frame_id=;
#
# print res['coor']
# break
# print sticks[0];
# videos=[vid for vid in os.listdir(path) if os.path.isdir(os.path.join(path,vid))];
# for vid in videos:
# path_vid = os.path.join(path,vid);
# shots=[shot for shot in os.listdir(path_vid) if os.path.isdir(os.path.join(path_vid,shot))];
# for shot in shots:
# path_shot = os.path.join(path_vid,shot);
return
meta_dir = '/disk2/januaryExperiments/tube_scoring'
out_file_html_pre = '/disk2/januaryExperiments/tube_scoring/best_tubes_comparison'
rel_path = ['/disk2', '../../..']
paths_to_im = [
os.path.join(meta_dir, 'images'),
os.path.join(meta_dir, 'images_perShot')
]
height_width = [500, 800]
columns = ['Frame Level', 'Shot Level']
for class_idx in range(10):
out_file_html = out_file_html_pre + '_' + str(class_idx) + '.html'
files = [
file_curr for file_curr in os.listdir(paths_to_im[0])
if file_curr.endswith('.png')
and file_curr.startswith(str(class_idx) + '_')
]
img_paths = []
captions = []
for file_curr in files:
img_paths_row = []
captions_row = []
for idx_path_to_im, path_to_im in enumerate(paths_to_im):
im_curr = os.path.join(path_to_im, file_curr).replace(
rel_path[0], rel_path[1])
img_paths_row.append(im_curr)
captions_row.append(columns[idx_path_to_im] + ' ' + file_curr)
img_paths.append(img_paths_row)
captions.append(captions_row)
visualize.writeHTML(out_file_html, img_paths, captions,
height_width[0], height_width[1])
return
params_dict = {}
params_dict['path_to_data'] = '/disk2/youtube/categories'
params_dict[
'score_dir'] = '/disk2/januaryExperiments/tube_scoring/scores_perShot'
params_dict['mat_dir'] = '/disk2/res11'
params_dict[
'out_dir'] = '/disk2/januaryExperiments/tube_scoring/images_perShot'
if not os.path.exists(params_dict['out_dir']):
os.mkdir(params_dict['out_dir'])
params_dict['to_exclude'] = [
'/disk2/res11/horse_7_16.mat', '/disk2/res11/horse_7_14.mat',
'/disk2/res11/horse_4_49.mat'
]
params_dict['class_labels_map'] = [('boat', 2), ('train', 9), ('dog', 6),
('cow', 5), ('aeroplane', 0),
('motorbike', 8), ('horse', 7),
('bird', 1), ('car', 3), ('cat', 4)]
params_dict['n_jobs'] = 12
params = createParams('drawTubeBB')
params = params(**params_dict)
script_drawTubeBB(params)
return
shot_dir = '/disk2/januaryExperiments/shot_score_normalized_perShot'
out_dir = '/disk2/januaryExperiments/tube_scoring/scores_perShot'
if not os.path.exists(out_dir):
os.mkdir(out_dir)
class_labels_map = [('boat', 2), ('train', 9), ('dog', 6), ('cow', 5),
('aeroplane', 0), ('motorbike', 8), ('horse', 7),
('bird', 1), ('car', 3), ('cat', 4)]
n_jobs = 12
[class_labels, class_idx_all] = zip(*class_labels_map)
check_list = [str(class_idx) for class_idx in class_idx_all]
# print check_list
file_list = [
file_curr for file_curr in os.listdir(shot_dir)
if file_curr.endswith('.p') and file_curr[0] in check_list
]
args = [(os.path.join(shot_dir,
file_curr), os.path.join(out_dir, file_curr), idx)
for idx, file_curr in enumerate(file_list)]
print len(args)
p = multiprocessing.Pool(n_jobs)
p.map(saveBestTubeAvgScore, args)
# for class_idx in class_idx_all:
# rel_files=[file_curr for file_curr in file_list if file_curr.startswith(str(class_idx)+'_') and file_curr.endswith('.p')];
# for rel_file in rel_files:
# score_file=os.path.join(shot_dir,rel_file);
# out_file=os.path.join(out_dir,rel_file);
# best_tube_rank,tubes_ranked,scores_ranked = getBestTubeAvgScore(score_file)
# pickle.dump([best_tube_rank,tubes_ranked,scores_ranked],open(out_file,'wb'));
# out_files=[os.path.join(shot_dir,file_curr) for file_curr in file_list if file_curr.startswith(str(class_idx)+'_') and file_curr.endswith('.p')];
# score_file=os.path.join(shot_dir,str(class_idx)+'_'+str(video_id)+'_'+str(shot_id)+'.p');
# best_tube_rank,tubes_ranked,scores_ranked = getBestTubeAvgScore(os.path.join(shot_dir,score_file))
return
path_to_data = '/disk2/youtube/categories'
to_exclude = [
'/disk2/res11/horse_7_16.mat', '/disk2/res11/horse_7_14.mat',
'/disk2/res11/horse_4_49.mat'
]
meta_dir = '/disk2/res11'
mat_files = pickle.load(open('/disk2/temp/horse_problem.p', 'rb'))
mat_files = [
os.path.join(meta_dir, file_curr[file_curr.rindex('/') + 1:] + '.mat')
for file_curr in mat_files
]
# for mat_file in mat_files:
mat_file = '/disk2/res11/horse_7_11.mat'
out_file = '/disk2/temp.png'
drawTubeBB(mat_file, path_to_data, out_file)
return
path_to_data = '/disk2/youtube/categories'
out_dir_patches = '/disk2/res11/tubePatches'
path_to_mat = '/disk2/res11'
if not os.path.exists(out_dir_patches):
os.mkdir(out_dir_patches)
mat_files = [
os.path.join(path_to_mat, file_curr)
for file_curr in os.listdir(path_to_mat) if file_curr.endswith('.mat')
]
mat_file = mat_files[0]
print mat_file
drawTubeBB(mat_file)
# script_saveTubePatches(mat_files,path_to_data,out_dir_patches,numThreads=8)
return
shot_dir = '/disk2/januaryExperiments/shot_score_normalized'
class_idx = 0
video_id = 1
shot_id = 1
score_file = os.path.join(
shot_dir,
str(class_idx) + '_' + str(video_id) + '_' + str(shot_id) + '.p')
best_tube_rank, tubes_ranked, scores_ranked = getBestTubeAvgScore(
os.path.join(shot_dir, score_file))
print best_tube_rank, tubes_ranked, scores_ranked