def writeSmallDatasetFile(out_file_pre,horse_data,num_neighbor,
num_data,in_file_horse,in_file_face,in_file_face_noIm,post_tags=None):
if post_tags is None:
post_tags=['_horse.txt','_face.txt','_face_noIm.txt'];
in_files=[in_file_horse,in_file_face,in_file_face_noIm];
data_org=util.readLinesFromFile(in_file_horse);
data_org=np.array(data_org);
idx_keep_all=[];
print horse_data.shape
horse_data=horse_data[:num_data];
for horse_curr in horse_data:
idx_curr=np.where(data_org==horse_curr)[0];
idx_curr=np.sort(idx_curr)
idx_keep=idx_curr[:num_neighbor];
idx_keep_all=idx_keep_all+list(idx_keep);
# print num_data,idx_keep
idx_keep_all=np.array(idx_keep_all);
print idx_keep_all.shape
files_to_return=[];
for idx_in_file,in_file in enumerate(in_files):
out_file_curr=out_file_pre+post_tags[idx_in_file];
if idx_in_file==0:
data_keep=data_org[idx_keep_all];
else:
data_curr=util.readLinesFromFile(in_file);
data_curr=np.array(data_curr);
data_keep=data_curr[idx_keep_all];
util.writeFile(out_file_curr,data_keep);
files_to_return.append(out_file_curr);
return files_to_return;
def makeCulpritFile():
out_dir = '/home/SSD3/maheen-data/temp/debug_problem_batch'
file_human = '/home/SSD3/maheen-data/horse_project/aflw/matches_5_train_fiveKP_noIm.txt'
file_horse = '/home/SSD3/maheen-data/horse_project/horse_resize/matches_5_train_fiveKP.txt'
new_file_human = file_human[:file_human.rindex('.')] + '_debug.txt'
new_file_horse = file_horse[:file_horse.rindex('.')] + '_debug.txt'
batch_no = 3
batch_size = 64
data_horse = util.readLinesFromFile(file_horse)
data_human = util.readLinesFromFile(file_human)
assert len(data_horse) == len(data_human)
print(len(data_horse) / batch_size)
# for batch_no in range(71,72):
batch_no = 71
line_idx = (batch_size * (batch_no - 1)) % len(data_horse)
print('____')
print(batch_no)
print(line_idx)
print data_horse[line_idx]
print data_human[line_idx]
data_horse_rel = data_horse[line_idx:line_idx + batch_size]
data_human_rel = data_human[line_idx:line_idx + batch_size]
assert len(data_horse_rel) == batch_size
assert len(data_human_rel) == batch_size
util.writeFile(new_file_horse, data_horse_rel)
util.writeFile(new_file_human, data_human_rel)
print new_file_human
print new_file_horse
def script_vizForHMDB():
out_dir='/disk2/mayExperiments/debug_finetuning/hmdb';
clusters_file='/home/maheenrashid/Downloads/debugging_jacob/optical_flow_prediction_test/examples/opticalflow/clusters.mat';
vid_list=os.path.join(out_dir,'video_list.txt');
out_dir_viz=os.path.join(out_dir,'im');
util.mkdir(out_dir_viz);
out_file_html=out_dir_viz+'.html';
path_to_hmdb='/disk2/marchExperiments/hmdb_try_2/hmdb'
dirs=util.readLinesFromFile(vid_list);
dirs=[os.path.join(path_to_hmdb,dir_curr) for dir_curr in dirs[2:]];
random.shuffle(dirs);
num_to_evaluate=100;
out_file_tif=os.path.join(out_dir,'tif_list.txt');
# recordContainingFiles(dirs,num_to_evaluate,out_file_flo,post_dir='images',ext='.flo');
tif_files=util.readLinesFromFile(out_file_tif);
tif_files=tif_files[:100];
img_files=[file_curr.replace('.tif','.jpg') for file_curr in tif_files];
flo_files=[file_curr.replace('.tif','.flo') for file_curr in tif_files];
clusters=po.readClustersFile(clusters_file);
script_writeFloVizHTML(out_file_html,out_dir_viz,flo_files,img_files,tif_files,clusters,True)
def writeMinLossFileLossData(out_file_pre,post_tags,minloss_post,loss_file):
new_files=[out_file_pre+post_tag_curr for post_tag_curr in post_tags];
horse_data=util.readLinesFromFile(new_files[0]);
horse_data=np.array(horse_data);
horse_data_uni=np.unique(horse_data);
face_data=util.readLinesFromFile(new_files[1]);
face_data_noIm=util.readLinesFromFile(new_files[2]);
assert len(face_data)==len(face_data_noIm);
loss_all=np.load(loss_file);
loss_all=loss_all[:len(face_data)];
assert loss_all.shape[0]==len(face_data);
new_data=[[],[],[]];
for idx_curr,horse_curr in enumerate(horse_data_uni):
idx_rel=np.where(horse_data==horse_curr)[0];
loss_rel=loss_all[idx_rel];
min_idx=np.argmin(loss_rel);
min_idx_big=idx_rel[min_idx];
assert loss_rel[min_idx]==loss_all[min_idx_big];
new_data[0].append(horse_curr);
new_data[1].append(face_data[min_idx_big]);
new_data[2].append(face_data_noIm[min_idx_big]);
new_files_out=[new_file_curr[:new_file_curr.rindex('.')]+minloss_post for new_file_curr in new_files];
for new_file_to_write,data_to_write in zip(new_files_out,new_data):
print new_file_to_write,len(data_to_write);
util.writeFile(new_file_to_write,data_to_write);
def main():
return
out_dir='/group/leegrp/maheen_data/flo_all_predictions_finetuned_model/results'
dir_flo_im='/group/leegrp/maheen_data/flo_all_predictions_finetuned_model/flo_npy';
util.mkdir(dir_flo_im);
h5_file_list='/group/leegrp/maheen_data/flo_all_predictions_finetuned_model/h5_file_list.txt';
util.mkdir(dir_flo_im);
clusters_file='/group/leegrp/maheen_data/flo_all_predictions_finetuned_model/clusters.mat';
vision3_path='/disk2/marchExperiments';
hpc_path='/group/leegrp/maheen_data';
# replace=['flo_all_predictions','flo_all_predictions_finetuned_model']
h5_files=util.readLinesFromFile(h5_file_list);
# h5_files=[file_curr.replace(replace[0],replace[1]) for file_curr in h5_files];
print h5_files[0];
# print len(h5_files);
# print os.path.exists(h5_files[0]);
# util.writeFile(h5_file_list,h5_files);
script_saveFloAsNpPred(clusters_file,h5_files,dir_flo_im,[vision3_path,hpc_path])
return
dir_meta_images='/group/leegrp/maheen_data/youtube';
flo_files_list='/group/leegrp/maheen_data/youtube_list_flo_paths.txt';
out_dir_numpy='/group/leegrp/maheen_data/youtube_resized_flo_npy_new';
util.mkdir(out_dir_numpy);
flo_files=util.readLinesFromFile(flo_files_list);
img_files=[];
args=[];
for idx,flo_file in enumerate(flo_files):
flo_name=flo_file[flo_file.rindex('/')+1:];
video_name=flo_name[:flo_name.index('.')];
just_name=flo_name[:flo_name.rindex('.')]
img_name=just_name+'.jpg';
img_path=os.path.join(dir_meta_images,video_name,'images_transfer',img_name);
out_file_num=os.path.join(out_dir_numpy,just_name+'.npy');
args.append((idx,flo_file,img_path,out_file_num));
print len(args)
print args[0];
# print multiprocessing.cpu_count()
p=multiprocessing.Pool(multiprocessing.cpu_count());
p.map(saveResizeFlo,args);
def parseAnnoFile(path_txt, path_pre=None, face=False):
face_data = util.readLinesFromFile(path_txt)
path_im = []
bbox = []
anno_points = []
for line_curr in face_data:
line_split = line_curr.split(' ')
pts = [float(str_curr) for str_curr in line_split[1:]]
pts = [int(str_curr) for str_curr in pts]
if path_pre is not None:
path_im.append(
os.path.join(path_pre, line_split[0].replace('\\', '/')))
else:
path_im.append(line_split[0])
bbox.append(pts[:4])
if face:
increment = 2
else:
increment = 3
anno_points_curr = [
pts[start:start + increment]
for start in range(4, len(pts), increment)
]
anno_points.append(anno_points_curr)
return path_im, bbox, anno_points
def main():
text_list='/disk2/aprilExperiments/dual_flow/list_of_dats_to_move.txt';
text_mv='/disk2/aprilExperiments/dual_flow/list_of_dats_to_move_commands.sh';
models=util.readLinesFromFile(text_list);
path_to_storage='/media/maheenrashid/Seagate\ Backup\ Plus\ Drive/maheen_data';
path_to_replace='/disk2';
mv_commands=[];
for model in models:
if not os.path.exists(model):
continue;
dir_curr=model[:model.rindex('/')];
dir_new=dir_curr.replace(path_to_replace,path_to_storage);
# print dir_new;
command='mkdir -p '+dir_new;
# print command;
mv_command='mv -v '+model+' '+dir_new+'/';
# print mv_command
mv_commands.append(mv_command);
subprocess.call(command,shell=True);
# raw_input();
util.writeFile(text_mv,mv_commands);
print text_mv
def script_saveHumanOnlyNeg():
out_file='/disk2/aprilExperiments/positives_160_human.txt'
out_dir='/disk2/aprilExperiments/negatives_npy_onlyHuman';
util.mkdir(out_dir);
im_pre='COCO_train2014_'
lines=util.readLinesFromFile(out_file);
img_files=[line[:line.index(' ')] for line in lines];
img_names=util.getFileNames(img_files,ext=False);
img_name=img_names[0];
print img_name
img_name_split=img_name.split('_');
idx_all=[int(img_name.split('_')[-1]) for img_name in img_names];
print len(img_names),len(idx_all),idx_all[0];
cat_id=1;
path_to_anno='/disk2/ms_coco/annotations';
anno_file='instances_train2014.json';
anno=json.load(open(os.path.join(path_to_anno,anno_file),'rb'))['annotations'];
script_saveBboxFiles(anno,out_dir,im_pre,idx_all,cat_id)
def shortenTrainingData(train_txt,train_txt_new,ratio_txt,val_txt_new=None):
# pos_human='/disk3/maheen_data/headC_160/noFlow_gaussian_human/pos_flos/positives_onlyHuman_withFlow.txt';
# neg_human='/disk3/maheen_data/headC_160/neg_flos/negatives_onlyHuman_withFlow.txt';
# pos_human_small='/disk3/maheen_data/headC_160/noFlow_gaussian_human/pos_flos/positives_onlyHuman_withFlow_oneHundreth.txt';
# neg_human_small='/disk3/maheen_data/headC_160/neg_flos/negatives_onlyHuman_withFlow_oneHundreth.txt';
# ratio_txt=100;
# shortenTrainingData(pos_human,pos_human_small,ratio_txt);
# shortenTrainingData(neg_human,neg_human_small,ratio_txt);
train_data=util.readLinesFromFile(train_txt);
# print ratio_txt
if ratio_txt<1:
ratio_txt=int(len(train_data)*ratio_txt);
# print ratio_txt;
random.shuffle(train_data);
train_data_new=train_data[:ratio_txt];
print len(train_data),len(train_data_new);
util.writeFile(train_txt_new,train_data_new);
if val_txt_new is not None:
val_data=train_data[ratio_txt:];
print len(val_data);
util.writeFile(val_txt_new,val_data);
def script_writeCommandsForPreprocessing(all_dirs_file,
command_file_pre,
num_proc,
check_file=None):
all_dirs = util.readLinesFromFile(all_dirs_file)
all_dirs = [dir_curr[:-1] for dir_curr in all_dirs]
if check_file is not None:
all_dirs = getRemainingDirs(all_dirs, check_file)
command_pre = 'echo '
command_middle_1 = ';cd ~/Downloads/opticalflow; matlab -nojvm -nodisplay -nosplash -r "out_folder=\''
command_middle = '\';saveTrainingData" > '
command_end = ' 2>&1'
commands = []
for dir_curr in all_dirs:
dir_curr = util.escapeString(dir_curr)
log_file = os.path.join(dir_curr, 'log.txt')
command = command_pre + dir_curr + command_middle_1 + dir_curr + command_middle + log_file + command_end
commands.append(command)
idx_range = util.getIdxRange(len(commands), len(commands) / num_proc)
command_files = []
for i, start_idx in enumerate(idx_range[:-1]):
command_file_curr = command_file_pre + str(i) + '.txt'
end_idx = idx_range[i + 1]
commands_rel = commands[start_idx:end_idx]
util.writeFile(command_file_curr, commands_rel)
command_files.append(command_file_curr)
return command_files
def saveGTData(path_to_txt, class_labels_map, out_file):
[class_labels, class_idx_all] = zip(*class_labels_map)
sticks = util.readLinesFromFile(path_to_txt)
print len(sticks)
meta_info = []
coords = []
for stick in sticks:
stick_split = stick.split('/')
stick_split = [curr for curr in stick_split if curr != '']
class_label = stick_split[3]
class_idx = class_idx_all[class_labels.index(class_label)]
video_id = int(stick_split[5])
shot_id = int(stick_split[7])
frame_id = stick_split[-1]
frame_id = int(frame_id[:frame_id.index('.')].strip('frame'))
meta_info_curr = [class_idx, video_id, shot_id, frame_id]
res = scipy.io.loadmat(stick, squeeze_me=True, struct_as_record=False)
boxes = res['coor']
if not hasattr(boxes[0], '__iter__'):
boxes = [boxes]
else:
print 'found it!', boxes
for box in boxes:
meta_info.append(meta_info_curr)
coords.append(list(box))
pickle.dump([meta_info, coords], open(out_file, 'wb'))
def addDimFaceData():
face_data_dir = '/home/SSD3/maheen-data/face_data/npy'
out_dir = '/home/SSD3/maheen-data/face_data/npy_dimAdd'
util.mkdir(out_dir)
face_data_file = os.path.join(face_data_dir, 'data_list.txt')
in_files = util.readLinesFromFile(face_data_file)
args = []
for idx, in_file in enumerate(in_files):
out_file = in_file.replace(face_data_dir, out_dir)
if os.path.exists(out_file):
continue
folder_curr = out_file[:out_file.rindex('/')]
util.mkdir(folder_curr)
args.append((idx, in_file, out_file))
print len(args)
# for arg in args:
# addDim(arg);
# break;
p = multiprocessing.Pool(multiprocessing.cpu_count())
p.map(addDim, args)
def saveRecordOfCountErrorFiles(score_files, class_label, path_to_patches,
out_file):
record = []
for idx_file_curr, file_curr in enumerate(score_files):
print idx_file_curr, len(score_files)
file_name = file_curr[file_curr.rindex('/') + 1:file_curr.rindex('.')]
class_idx = int(file_name[:file_name.index('_')])
video_id = int(file_name[file_name.index('_') +
1:file_name.rindex('_')])
shot_id = int(file_name[file_name.rindex('_') + 1:])
scores = pickle.load(open(file_curr, 'rb'))
shot_string = class_label
# class_labels[class_idx_all.index(class_idx)];
shot_string = os.path.join(
path_to_patches,
shot_string + '_' + str(video_id) + '_' + str(shot_id))
for tube_id in scores:
scores_curr = scores[tube_id]
txt_file = os.path.join(path_to_patches, shot_string, str(tube_id),
str(tube_id) + '.txt')
patch_paths = util.readLinesFromFile(txt_file)
if scores_curr.shape[0] != len(patch_paths):
print 'PROBLEM'
record.append(
(txt_file, file_curr, tube_id, scores_curr.shape[0],
len(patch_paths)))
print len(record)
pickle.dump(record, open(out_file, 'wb'))
def getListScoresAndPatches_multiProc((idx_file_curr, file_curr, class_label,
path_to_patches)):
list_scores = []
list_files = []
print idx_file_curr, file_curr
file_name = file_curr[file_curr.rindex('/') + 1:file_curr.rindex('.')]
class_idx = int(file_name[:file_name.index('_')])
video_id = int(file_name[file_name.index('_') + 1:file_name.rindex('_')])
shot_id = int(file_name[file_name.rindex('_') + 1:])
scores = pickle.load(open(file_curr, 'rb'))
shot_string = class_label
shot_string = os.path.join(
path_to_patches,
shot_string + '_' + str(video_id) + '_' + str(shot_id))
for tube_id in scores:
scores_curr = scores[tube_id]
scores_curr = np.mean(scores_curr, axis=1)
list_scores.extend(list(scores_curr))
txt_file = os.path.join(path_to_patches, shot_string, str(tube_id),
str(tube_id) + '.txt')
patch_paths = util.readLinesFromFile(txt_file)
assert len(patch_paths) == scores_curr.shape[0]
list_files.extend(patch_paths)
return (list_scores, list_files)
def experiment_nnPatches(params):
out_file_text=params.out_file_text;
class_ids=params.class_id;
img_paths=util.readLinesFromFile(out_file_text)
class_id_idx_tuples=[];
for img_path in img_paths:
class_id=img_path[:img_path.rindex('_')]
class_id=class_id[class_id.rindex('_')+1:];
class_idx=class_ids.index(class_id);
class_idx=params.class_idx[class_idx];
class_id_idx_tuples.append((class_id,class_idx));
file_names_mat,object_indices=recreateOriginalPaths(params.path_to_annotation,img_paths,returnObject_idx=True);
print 'getting azimuths'
azimuths=[getObjectStruct(file_name,object_idx).viewpoint.azimuth_coarse for file_name,object_idx in zip(file_names_mat,object_indices)]
if params.out_file_layers is None:
print 'running layers part'
out_file_layers=caffe_wrapper.saveFeaturesOfLayers(out_file_text,params.path_to_classify,params.gpu_no,params.layers,ext='jpg',out_file=params.out_file_pre,meanFile=params.caffe_mean,deployFile=params.caffe_deploy,modelFile=params.caffe_model,images_dim=params.images_dim)
params=params._replace(out_file_layers=out_file_layers)
out_file_layers=params.out_file_layers;
print 'writing to db'
for layer in params.layers:
vals=np.load(out_file_layers);
indices,distances=nearest_neighbor.doCosineDistanceNN(vals[layer],numberOfN=None);
mani=Pascal3D_Manipulator(params.db_path_out);
mani.openSession();
for idx in range(len(img_paths)):
mani.insert(idx,img_paths[idx],layer,out_file_layers,class_id_idx_tuples[idx][0],class_id_idx_tuples[idx][1],params.caffe_model, azimuth=azimuths[idx],neighbor_index=indices[idx],neighbor_distance=distances[idx],trainedClass=params.trainFlag,commitFlag=False)
mani.closeSession();
return params
def script_writeCommandsForPreprocessing(all_dirs_file,command_file_pre,num_proc,check_file=None):
all_dirs=util.readLinesFromFile(all_dirs_file);
all_dirs=[dir_curr[:-1] for dir_curr in all_dirs];
if check_file is not None:
all_dirs=getRemainingDirs(all_dirs,check_file);
command_pre='echo '
command_middle_1=';cd ~/Downloads/opticalflow; matlab -nojvm -nodisplay -nosplash -r "out_folder=\''
command_middle='\';saveTrainingData" > '
command_end=' 2>&1';
commands=[];
for dir_curr in all_dirs:
dir_curr=util.escapeString(dir_curr);
log_file=os.path.join(dir_curr,'log.txt');
command=command_pre+dir_curr+command_middle_1+dir_curr+command_middle+log_file+command_end;
commands.append(command);
idx_range=util.getIdxRange(len(commands),len(commands)/num_proc)
command_files=[];
for i,start_idx in enumerate(idx_range[:-1]):
command_file_curr=command_file_pre+str(i)+'.txt'
end_idx=idx_range[i+1]
commands_rel=commands[start_idx:end_idx];
util.writeFile(command_file_curr,commands_rel);
command_files.append(command_file_curr);
return command_files;
def script_saveMatFiles(flo_dir,im_dir,out_dir,mat_file,proto_file):
#get video name
video_name=flo_dir[:-1];
video_name=video_name[video_name.rindex('/')+1:];
print video_name
#get flo files
flo_files=[os.path.join(flo_dir,file_curr) for file_curr in os.listdir(flo_dir) if file_curr.endswith('.flo')];
flo_files.sort();
#get im files
im_files=util.readLinesFromFile(os.path.join(flo_dir,'im_1.txt'));
old_dir=im_files[0][:im_files[0].rindex('/')+1];
#if dirs have changed, replace the paths
if im_dir!=old_dir:
im_files=[im_curr.replace(old_dir,im_dir) for im_curr in im_files];
#get batch size
batch_size=getBatchSizeFromDeploy(os.path.join(flo_dir,proto_file));
#get batch info
batch_num=[int(file_curr[file_curr.rindex('-')+1:file_curr.rindex('(')]) for file_curr in flo_files];
batch_num=np.array(batch_num);
batch_ids=list(set(batch_num))
batch_ids.sort();
flo_files_all = [];
im_files_all = []
for batch_no in batch_ids:
idx_rel=np.where(batch_num==batch_no)[0];
flo_files_curr=[];
im_files_curr=[];
for idx_curr in idx_rel:
flo_file=flo_files[idx_curr];
im_no=int(flo_file[flo_file.rindex('(')+1:flo_file.rindex(')')]);
im_corr=im_files[batch_no*batch_size+im_no];
flo_files_curr.append(flo_file);
im_files_curr.append(im_corr);
flo_files_all.append(flo_files_curr);
im_files_all.append(im_files_curr);
#save as mat with flofiles, im_files, and out_dir;
for idx_batch_no,batch_no in enumerate(batch_ids):
flo_files=flo_files_all[idx_batch_no];
im_files=im_files_all[idx_batch_no];
out_dir_mat = os.path.join(out_dir,video_name+'_'+str(batch_no));
# print out_dir_mat
if not os.path.exists(out_dir_mat):
os.mkdir(out_dir_mat);
out_file=os.path.join(out_dir_mat,mat_file);
print out_file
mat_data={'flo_files':flo_files,'im_files':im_files}
scipy.io.savemat(out_file,mat_data)
def readPredFile(pred_file):
lines = util.readLinesFromFile(pred_file)
annos_all = []
for line_curr in lines:
annos_str = line_curr.split()
annos = parseAnnoStr(annos_str)
annos_all.append(annos)
return annos_all
def script_writeFlownetCommands(params):
video_list_file = params.video_list_file
path_to_video_meta = params.path_to_video_meta
in_dir_meta = params.in_dir_meta
out_dir_meta = params.out_dir_meta
path_to_deploy = params.path_to_deploy
out_file_commands = params.out_file_commands
dir_flownet_meta = params.dir_flownet_meta
path_to_sizer = params.path_to_sizer
caffe_bin = params.caffe_bin
path_to_model = params.path_to_model
text_1_org = params.text_1
text_2_org = params.text_2
deploy_file = params.deploy_file
gpu = params.gpu
im_dirs = util.readLinesFromFile(video_list_file)
im_dirs = [
im_dir.replace(path_to_video_meta, in_dir_meta)[:-4]
for im_dir in im_dirs
]
commands = []
# im_dirs=['/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/image_data/hmdb/pick/THE_WALLET_TRICK!!!_pick_f_cm_np2_ba_med_1'];
for idx_im_dir, im_dir in enumerate(im_dirs):
print idx_im_dir, len(im_dirs)
out_dir_curr = im_dir.replace(in_dir_meta, out_dir_meta)
text_1 = os.path.join(out_dir_curr, text_1_org)
text_2 = os.path.join(out_dir_curr, text_2_org)
out_deploy = os.path.join(out_dir_curr, deploy_file)
subprocess.call('mkdir -p ' + util.escapeString(out_dir_curr),
shell=True)
list_1, list_2 = getImageListForFlow(im_dir)
util.writeFile(text_1, list_1)
util.writeFile(text_2, list_2)
# im_test=util.escapeString(list_1[0]);
dim_list = [
int(dimstr) for dimstr in str(
subprocess.check_output([path_to_sizer, list_1[0]])).split(',')
]
replaceProto(path_to_deploy, out_deploy, dim_list, text_1, text_2,
len(list_1), out_dir_curr)
args = [
caffe_bin, 'test', '-model',
util.escapeString(out_deploy), '-weights', path_to_model,
'-iterations', '1', '-gpu',
str(gpu)
]
cmd = str.join(' ', args)
commands.append(cmd)
# print('Executing %s' % cmd)
util.writeFile(out_file_commands, commands)
def convertFileToFloOnly(neg_flo,out_file_neg):
neg_flo=util.readLinesFromFile(neg_flo);
neg_only_flo=[];
for neg_flo_curr in neg_flo:
neg_flo_curr=neg_flo_curr.split(' ');
neg_only_flo.append(neg_flo_curr[-1]+' '+neg_flo_curr[1]);
assert len(neg_only_flo)==len(neg_flo);
util.writeFile(out_file_neg,neg_only_flo);
def getNumNeighbors(file_horse):
lines=util.readLinesFromFile(file_horse);
lines=np.array(lines);
uni_lines=np.unique(lines);
counts=np.zeros(uni_lines.shape);
for idx_uni_curr,uni_curr in enumerate(uni_lines):
if idx_uni_curr%500==0:
print idx_uni_curr;
counts[idx_uni_curr]=np.sum(lines==uni_curr);
return uni_lines,counts;
def writeTrainFilesWithFlow(old_train_file,dir_flo_im,new_train_file,ext='.png'):
lines=util.readLinesFromFile(old_train_file);
img_files=[line[:line.index(' ')] for line in lines];
file_names=util.getFileNames(img_files,ext=False);
flo_im_files=[os.path.join(dir_flo_im,file_name+ext) for file_name in file_names];
for flo_im_file in flo_im_files:
assert os.path.exists(flo_im_file);
lines_new=[line+' '+flo_im_curr for line,flo_im_curr in zip(lines,flo_im_files)];
util.writeFile(new_train_file,lines_new);
def main():
data_dir = '../../../../data/ck_96/train_test_files'
out_dir = '../../../../data/ck_96/train_test_files_tfrecords'
im_path_prepend = '../../../'
util.mkdir(out_dir)
split_num = 0
train_file = os.path.join(data_dir, 'train_' + str(split_num) + '.txt')
test_file = os.path.join(data_dir, 'test_' + str(split_num) + '.txt')
mean_file = os.path.join(data_dir, 'train_' + str(split_num) + '_mean.png')
std_file = os.path.join(data_dir, 'train_' + str(split_num) + '_std.png')
mean_im = scipy.misc.imread(mean_file).astype(np.float32)
std_im = scipy.misc.imread(std_file).astype(np.float32)
std_im[std_im == 0] = 1
for in_file in [train_file, test_file]:
out_file = os.path.join(
out_dir,
os.path.split(in_file)[1].replace('.txt', '.tfrecords'))
print in_file, out_file
lines = util.readLinesFromFile(in_file)
random.shuffle(lines)
writer = tf.python_io.TFRecordWriter(out_file)
for idx_line, line in enumerate(lines):
if idx_line % 100 == 0:
print idx_line, line
im_path, label = line.split(' ')
label = int(label)
im_path = os.path.join(im_path_prepend, im_path)
img = load_image(im_path, mean_im, std_im)
# print img.shape,np.min(img),np.max(img)
feature = {
'image_raw':
_bytes_feature(tf.compat.as_bytes(img.tostring())),
'label': _int64_feature(label),
'height': _int64_feature(img.shape[0]),
'width': _int64_feature(img.shape[1]),
'depth': _int64_feature(img.shape[2])
}
example = tf.train.Example(features=tf.train.Features(
feature=feature))
# Serialize to string and write on the file
writer.write(example.SerializeToString())
writer.close()
def main():
pos_data_txt='/disk2/februaryExperiments/deep_proposals/positive_data.txt';
num_to_keep=100;
out_file_db='/disk2/februaryExperiments/deep_proposals/positive_data_'+str(num_to_keep)+'.hdf5';
lines=util.readLinesFromFile(pos_data_txt);
ims=[];
masks=[];
for line in lines:
idx_space=line.index(' ');
im=line[:idx_space];
mask=line[idx_space+1:];
ims.append(im);
masks.append(mask);
if num_to_keep is None:
num_to_keep=len(ims);
ims=ims[:num_to_keep];
masks=masks[:num_to_keep];
f=h5py.File(out_file_db,'w')
canoncial_shape=(240,240);
dset = f.create_dataset('images', (len(ims),3,canoncial_shape[0],canoncial_shape[1]), dtype='i')
dset = f.create_dataset('masks', (len(masks),1,canoncial_shape[0],canoncial_shape[1]), dtype='i')
for idx in range(len(ims)):
print idx
img=scipy.misc.imread(ims[idx]);
# print img.shape
if len(img.shape)<3:
img=np.dstack((img,img,img))
# print img.shape
f['/images'][idx,...]=reshapeForDB(img,canoncial_shape,addDim=False);
f['/masks'][idx,...]=reshapeForDB(scipy.misc.imread(masks[idx]),canoncial_shape,addDim=True);
# print np.min(im_curr),np.max(im_curr);
# print im_curr.shape
# mask_curr=scipy.misc.imread(masks[idx]);
# mask_curr=np.expand_dims(mask_curr,3)
# mask_curr=np.transpose(mask_curr,(2,0,1));
# print np.min(mask_curr),np.max(mask_curr);
# print mask_curr.shape
# f['/images'][idx,...]=im_curr;
# f['/masks'][idx,...]=mask_curr;
f.close();
print len(ims);
print ims[:3];
print len(masks);
print masks[:3];
def writeTrainValTxtExcluded(train_new_text,val_new_text,out_file_text,training_data_text,percent_exclude):
lines=util.readLinesFromFile(out_file_text);
info=[tuple(line_curr.split(' ')) for line_curr in lines];
class_rec={};
for dataset,video, in info:
if dataset=='youtube':
video_split=video.split('_');
class_curr=video_split[0];
if class_curr in class_rec:
class_rec[class_curr].append(video);
else:
class_rec[class_curr]=[video];
list_exclude_all=[];
for class_curr in class_rec.keys():
num_exclude=int(math.ceil(len(class_rec[class_curr])*percent_exclude));
list_shuffle=class_rec[class_curr];
random.shuffle(list_shuffle);
list_exclude=list_shuffle[:num_exclude];
list_exclude_all=list_exclude_all+list_exclude;
lines=util.readLinesFromFile(training_data_text);
# print len(lines);
lines_to_keep=[];
lines_to_exclude=[];
for line in lines:
img=line[:line.index(' ')];
img_split=img.split('/');
if img_split[3]=='youtube' and (img_split[4] in list_exclude_all):
lines_to_exclude.append(line);
# print img
continue;
else:
lines_to_keep.append(line);
print len(lines_to_keep),len(lines_to_exclude),len(lines),len(lines_to_keep)+len(lines_to_exclude)
util.writeFile(train_new_text,lines_to_keep);
util.writeFile(val_new_text,lines_to_exclude);
def getCommandFaceTest(path_to_th,out_dir,file_curr,batch_size=100):
command=['th',path_to_th];
command=command+['-val_data_path',file_curr];
command=command+['-outDir',out_dir];
command=command+['-batchSize',str(batch_size)];
amount_data=len(util.readLinesFromFile(file_curr));
num_iterations=amount_data/batch_size;
if amount_data%batch_size!=0:
num_iterations=num_iterations+1;
command=command+['-iterations',str(num_iterations)];
command=' '.join(command);
return command;
def writeh5ImgFile(dir_neg,out_file_match):
lines=[];
h5_files=[os.path.join(dir_neg,file_curr) for file_curr in os.listdir(dir_neg) if file_curr.endswith('.h5')];
print len(h5_files)
for idx_file_curr,file_curr in enumerate(h5_files):
if idx_file_curr%100==0:
print idx_file_curr
img_file=util.readLinesFromFile(file_curr.replace('.h5','.txt'))[0].strip();
# print file_curr,img_file
lines.append(file_curr+' '+img_file);
util.writeFile(out_file_match,lines);
def script_writeNegFile():
dir_flow='/disk2/aprilExperiments/deep_proposals/flow/results_neg'
out_text='/disk2/aprilExperiments/deep_proposals/flow/test_neg.txt';
# util.mkdir(dir_flow);
neg_text='/disk2/marchExperiments/deep_proposals/negatives.txt';
lines=util.readLinesFromFile(neg_text);
neg_images=[line_curr[:line_curr.index(' ')] for line_curr in lines];
neg_images=neg_images[:100];
to_write=[neg_image+' 1' for neg_image in neg_images]
util.writeFile(out_text,to_write);
def getDataAndLabels(data_path, batchSize, numBatches, fc6_dir):
lines = util.readLinesFromFile(data_path)
out_data = np.zeros((len(lines), 4096))
start_curr = 0
for i in range(numBatches):
file_curr = os.path.join(fc6_dir,
str(i + 1) + '.npy')
fc6_curr = np.load(file_curr)
end_curr = min(start_curr + fc6_curr.shape[0], len(lines))
len_curr = end_curr - start_curr
out_data[start_curr:end_curr, :] = fc6_curr[:len_curr]
start_curr = end_curr
return out_data, lines
def main(): # out_dir_meta='/disk2/mayExperiments/validation'; # val_file=os.path.join(out_dir_meta,'val.txt'); out_dir_meta='/disk2/mayExperiments/train_data'; val_file=os.path.join(out_dir_meta,'train.txt'); out_dir=os.path.join(out_dir_meta,'rescaled_images'); # util.mkdir(out_dir); img_paths=util.readLinesFromFile(val_file); # print len(img_paths); # img_paths=img_paths[:3]; rescaleImAndSaveMeta(img_paths,out_dir);
def saveFloAsNpPred((h5_file,dir_flo_im,replace_paths,C,idx)):
print idx;
img_file=util.readLinesFromFile(h5_file.replace('.h5','.txt'))[0].strip();
img_file=img_file.replace(replace_paths[0],replace_paths[1]);
img_name=img_file[img_file.rindex('/')+1:img_file.rindex('.')];
out_file_flo=os.path.join(dir_flo_im,img_name+'.npy');
if os.path.exists(out_file_flo):
return;
flo=getFlowMat(h5_file,C);
im=scipy.misc.imread(img_file);
flo_resize=resizeSP(flo,im.shape);
np.save(out_file_flo,flo_resize);
def main():
dir_server='/home/SSD3/maheen-data';
click_str='http://vision1.idav.ucdavis.edu:1000';
out_dir_meta_data='/home/SSD3/maheen-data/horse_project/data_check';
dir_neighbors='/home/SSD3/maheen-data/horse_project/neighbor_data';
matches_file=os.path.join(dir_neighbors,'horse_trainImageList_2_data_100_neigbors.txt');
out_dir_debug=os.path.join(dir_neighbors,'debug');
out_dir_breakdowns=os.path.join(dir_neighbors,'small_datasets');
file_pre='matches';
util.mkdir(out_dir_breakdowns);
util.mkdir(out_dir_debug);
out_file_counts=os.path.join(out_dir_debug,'counts.npz');
out_file_dist=os.path.join(out_dir_debug,'counts_dist.png');
out_dir_meta_horse = os.path.join(out_dir_meta_data,'horse');
out_dir_meta_face = os.path.join(out_dir_meta_data,'aflw');
num_neighbors=100;
out_file_face=os.path.join(out_dir_meta_face,'matches_'+str(num_neighbors)+'_train_allKP.txt');
out_file_face_noIm=os.path.join(out_dir_meta_face,'matches_'+str(num_neighbors)+'_train_allKP_noIm.txt');
out_file_horse=os.path.join(out_dir_meta_horse,'matches_'+str(num_neighbors)+'_train_allKP.txt');
old_horse_file='/home/SSD3/maheen-data/horse_project/data_check/horse/matches_5_train_allKP_minLoss_clean_full.txt';
old_data=util.readLinesFromFile(old_horse_file);
old_data=np.array(old_data);
old_data=np.unique(old_data);
# new_data,counts=getNumNeighbors(out_file_horse);
# np.savez(out_file_counts,new_data=new_data,counts=counts);
data=np.load(out_file_counts);
print data.files;
new_data=data['new_data'];
counts=data['counts'];
num_data=range(500,len(old_data),500);
num_data[-1]=len(old_data);
num_neighbors=range(5,25,5);
np.random.shuffle(old_data);
for num_neighbor in num_neighbors:
for num_data_curr in num_data:
file_curr=file_pre+'_'+str(num_neighbor)+'_'+str(num_data_curr);
out_file_pre=os.path.join(out_dir_breakdowns,file_curr);
files=writeSmallDatasetFile(out_file_pre,old_data,num_neighbor,num_data_curr,out_file_horse,\
out_file_face,out_file_face_noIm);
for file_curr in files:
print file_curr;
def script_testOnYoutube():
val_file='/disk2/mayExperiments/finetuning_youtube_hmdb_llr/val_eq.txt'
out_dir='/disk2/mayExperiments/eval_ucf_finetune';
clusters_file='/home/maheenrashid/Downloads/debugging_jacob/optical_flow_prediction_test/examples/opticalflow/clusters.mat';
gpu=0;
util.mkdir(out_dir);
# model_file='/home/maheenrashid/Downloads/debugging_jacob/optical_flow_prediction_test/examples/opticalflow/final.caffemodel';
# out_dir_model=os.path.join(out_dir,'original_model');
model_file='/disk2/mayExperiments/ft_youtube_hmdb_ucfClusters/OptFlow_youtube_hmdb__iter_55000.caffemodel';
out_dir_model=os.path.join(out_dir,'ft_ucf_model');
util.mkdir(out_dir_model);
out_dir_flo=os.path.join(out_dir_model,'flo');
out_dir_flo_viz=os.path.join(out_dir_model,'flo_viz');
util.mkdir(out_dir_flo);util.mkdir(out_dir_flo_viz)
num_to_pick=20;
img_paths=util.readLinesFromFile(val_file);
img_paths=[img_path[:img_path.index(' ')] for img_path in img_paths];
class_names=[file_curr[:file_curr.index('_')] for file_curr in util.getFileNames(img_paths)];
classes=list(set(class_names));
class_names=np.array(class_names);
img_paths_test=[];
for class_curr in classes:
idx_rel=np.where(class_names==class_curr)[0];
idx_rel=idx_rel[:num_to_pick];
img_paths_test.extend([img_paths[idx_curr] for idx_curr in idx_rel]);
# po.script_saveFlosAndViz(img_paths_test,out_dir_flo,out_dir_flo_viz,gpu,model_file,clusters_file);
out_file_html=os.path.join(out_dir,'model_comparison.html');
out_dirs_flo_viz=[os.path.join(out_dir,'original_model','flo_viz'),os.path.join(out_dir,'ft_ucf_model','flo_viz')];
out_dirs_flo_viz_captions=['original_model','ft_ucf_model'];
img_paths_html=[];
captions_html=[];
img_names=util.getFileNames(img_paths_test,ext=False);
for img_path_test,img_name in zip(img_paths_test,img_names):
row_curr=[];
row_curr.append(util.getRelPath(img_path_test));
for out_dir_curr in out_dirs_flo_viz:
file_curr=os.path.join(out_dir_curr,img_name+'.png');
row_curr.append(util.getRelPath(file_curr));
captions_curr=[img_name]+out_dirs_flo_viz_captions;
img_paths_html.append(row_curr)
captions_html.append(captions_curr);
visualize.writeHTML(out_file_html,img_paths_html,captions_html);
def getMovedDirPath(moved_dir, orig_dir, sub_dirs_file_name):
sub_dirs = [
util.readLinesFromFile(os.path.join(dir_curr, sub_dirs_file_name))
for dir_curr in [moved_dir, orig_dir]
]
full_paths = [
os.path.join(orig_dir, dir_curr) for dir_curr in sub_dirs[1]
if dir_curr not in sub_dirs[0]
]
full_paths = full_paths + [
os.path.join(moved_dir, dir_curr) for dir_curr in sub_dirs[0]
]
list_intersection = set(sub_dirs[0] + sub_dirs[1])
return full_paths
def saveMinEqualFrames(train_new_text,out_file_idx,out_file_eq,includeHuman=True):
lines=util.readLinesFromFile(train_new_text);
img_paths=[line[:line.index(' ')] for line in lines];
p=multiprocessing.Pool(multiprocessing.cpu_count());
vals=p.map(getDataSetAndVideoName,img_paths);
[dataset,video]=zip(*vals)
dataset=np.array(dataset);
print np.unique(dataset);
frame_idx_rec={};
if includeHuman:
frame_idx_rec['human']=list(np.where(dataset=='hmdb_try_2')[0]);
for idx,video_curr in enumerate(video):
if dataset[idx]=='youtube':
class_curr=video_curr[:video_curr.index('_')];
if class_curr in frame_idx_rec:
frame_idx_rec[class_curr].append(idx);
else:
frame_idx_rec[class_curr]=[idx];
for class_curr in frame_idx_rec.keys():
print class_curr,len(frame_idx_rec[class_curr]);
min_frames=min([len(val_curr) for val_curr in frame_idx_rec.values()]);
print 'min_frames',min_frames
idx_to_pick=[];
for class_curr in frame_idx_rec.keys():
idx_curr=frame_idx_rec[class_curr];
random.shuffle(idx_curr);
idx_to_pick.extend(idx_curr[:min_frames]);
# print class_curr,len(frame_idx_rec[class_curr]);
idx_all=[idx_curr for idx_curr_all in frame_idx_rec.values() for idx_curr in idx_curr_all];
print len(idx_all),len(lines);
assert len(idx_all)==len(lines);
idx_all.sort();
print idx_all==list(range(len(lines)));
assert idx_all==list(range(len(lines)));
lines_to_keep=[lines[idx_curr] for idx_curr in idx_to_pick];
print len(lines_to_keep);
np.save(out_file_idx,np.array(idx_to_pick))
util.writeFile(out_file_eq,lines_to_keep);
def script_writeHumanOnlyNegFile():
neg_file_old='/disk2/marchExperiments/deep_proposals/negatives.txt'
neg_file_new='/disk2/marchExperiments/deep_proposals/negatives_onlyHuman.txt'
npy_dir_old='/disk2/marchExperiments/deep_proposals/negatives'
npy_dir_new='/disk2/aprilExperiments/negatives_npy_onlyHuman'
lines=util.readLinesFromFile(neg_file_old);
lines_new=[line.replace(npy_dir_old,npy_dir_new) for line in lines];
for line in lines_new:
assert npy_dir_new in line;
print len(lines),len(lines_new);
print lines_new[0];
util.writeFile(neg_file_new,lines_new);
def getOutputInfoMP((list_file,out_files_test)):
img_file = util.readLinesFromFile(list_file.replace('.h5','.txt'))[0].strip();
if out_files_test is not None and img_file not in out_files_test:
line_curr=None;
else:
im=scipy.misc.imread(img_file);
if len(im.shape)>2:
str_size=[im.shape[0],im.shape[1],im.shape[2]];
else:
str_size=[im.shape[0],im.shape[1],1];
str_size=[str(i) for i in str_size]
line_curr=[list_file,img_file]+str_size;
line_curr=' '.join(line_curr);
return line_curr;
def script_visualizeFlos(params):
out_file_html = params.out_file_html
out_dir_flo_im = params.out_dir_flo_im
flo_dir = params.flo_dir
im_file_1 = params.im_file_1
im_file_2 = params.im_file_2
height = params.height
width = params.width
rel_path = params.rel_path
list_1=util.readLinesFromFile(im_file_1);
list_2=util.readLinesFromFile(im_file_2);
flo_files=[os.path.join(flo_dir,file_curr) for file_curr in os.listdir(flo_dir) if file_curr.endswith('.flo')];
flo_nums=[int(file_curr[file_curr.rindex('(')+1:file_curr.rindex(')')]) for file_curr in flo_files];
print len(list_1),len(list_2),len(flo_files);
flo_files_sorted=[];
for idx_flo in range(len(flo_nums)):
idx_curr=flo_nums.index(idx_flo);
flo_files_sorted.append(flo_files[idx_curr]);
im_list=[];
caption_list=[];
# for idx_flo,flo_file_curr in enumerate(flo_files_sorted):
for idx_flo in range(len(list_1)):
flo_file_curr=flo_files_sorted[idx_flo];
out_flo_x=os.path.join(out_dir_flo_im,str(idx_flo)+'_x.png');
out_flo_y=os.path.join(out_dir_flo_im,str(idx_flo)+'_y.png');
flo=util.readFlowFile(flo_file_curr);
visualize.visualizeFlo(flo,out_flo_x,out_flo_y);
im_file_rel_1=list_1[idx_flo].replace(rel_path[0],rel_path[1]);
im_file_rel_2=list_2[idx_flo].replace(rel_path[0],rel_path[1]);
flo_file_rel_1=out_flo_x.replace(rel_path[0],rel_path[1]);
flo_file_rel_2=out_flo_y.replace(rel_path[0],rel_path[1]);
im_list.append([im_file_rel_1,im_file_rel_2,flo_file_rel_1,flo_file_rel_2]);
caption_list.append(['Image 1','Image 2','Flow x','Flow y']);
visualize.writeHTML(out_file_html,im_list,caption_list,height=height,width=width)
def writeTrainingDataFiles(dir_content_file,pre_dir,img_dir,out_file_text,ignore_amount=-2,postfix='.jpg'):
start_idx=len(pre_dir);
files=util.readLinesFromFile(dir_content_file);
lines_to_write=[];
for idx_file_curr,file_curr in enumerate(files):
if idx_file_curr%1000==0:
print idx_file_curr
file_name=file_curr[start_idx+1:];
file_name=file_name.split('_');
file_name='_'.join(file_name[:ignore_amount]);
file_name=file_name+postfix;
file_name=os.path.join(img_dir,file_name);
lines_to_write.append(file_name+' '+file_curr);
util.writeFile(out_file_text,lines_to_write);
def writeMinLossFile(out_file_pre,post_tags,minloss_post,old_horse_file,old_human_file,old_human_file_noIm):
new_files=[out_file_pre+post_tag_curr for post_tag_curr in post_tags];
old_data=util.readLinesFromFile(old_horse_file);
old_data=np.array(old_data);
new_data=util.readLinesFromFile(new_files[0]);
new_data=np.array(new_data);
new_data_uni=np.unique(new_data);
bin_keep=np.in1d(old_data,new_data_uni);
# print bin_keep.shape,sum(bin_keep);
old_files=[old_horse_file,old_human_file,old_human_file_noIm];
new_files_write=[file_curr[:file_curr.rindex('.')]+minloss_post for file_curr in new_files];
for old_file_curr,new_file_curr,new_file_org in zip(old_files,new_files_write,new_files):
data_curr=util.readLinesFromFile(old_file_curr);
data_curr=np.array(data_curr);
data_keep=data_curr[bin_keep];
print old_file_curr,new_file_curr,len(data_keep);
print data_keep[0];
new_file_org_data=util.readLinesFromFile(new_file_org);
new_file_org_data=np.array(new_file_org_data);
bin_check=np.in1d(data_keep,new_file_org_data);
print sum(bin_check),data_keep.shape[0];
assert sum(bin_check)==data_keep.shape[0];
def script_writeFlownetCommands(params):
video_list_file= params.video_list_file;
path_to_video_meta= params.path_to_video_meta;
in_dir_meta= params.in_dir_meta;
out_dir_meta= params.out_dir_meta;
path_to_deploy= params.path_to_deploy;
out_file_commands= params.out_file_commands;
dir_flownet_meta= params.dir_flownet_meta;
path_to_sizer= params.path_to_sizer;
caffe_bin = params.caffe_bin;
path_to_model = params.path_to_model;
text_1_org= params.text_1;
text_2_org= params.text_2;
deploy_file= params.deploy_file;
gpu= params.gpu;
im_dirs=util.readLinesFromFile(video_list_file);
im_dirs=[im_dir.replace(path_to_video_meta,in_dir_meta)[:-4] for im_dir in im_dirs];
commands=[];
# im_dirs=['/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/image_data/hmdb/pick/THE_WALLET_TRICK!!!_pick_f_cm_np2_ba_med_1'];
for idx_im_dir,im_dir in enumerate(im_dirs):
print idx_im_dir,len(im_dirs);
out_dir_curr=im_dir.replace(in_dir_meta,out_dir_meta);
text_1=os.path.join(out_dir_curr,text_1_org);
text_2=os.path.join(out_dir_curr,text_2_org);
out_deploy=os.path.join(out_dir_curr,deploy_file);
subprocess.call('mkdir -p '+util.escapeString(out_dir_curr), shell=True)
list_1,list_2 = getImageListForFlow(im_dir)
util.writeFile(text_1,list_1);
util.writeFile(text_2,list_2);
# im_test=util.escapeString(list_1[0]);
dim_list = [int(dimstr) for dimstr in str(subprocess.check_output([path_to_sizer, list_1[0]])).split(',')]
replaceProto(path_to_deploy,out_deploy,dim_list,text_1,text_2,len(list_1),out_dir_curr)
args = [caffe_bin, 'test', '-model', util.escapeString(out_deploy),
'-weights', path_to_model,
'-iterations', '1',
'-gpu', str(gpu)]
cmd = str.join(' ', args)
commands.append(cmd);
# print('Executing %s' % cmd)
util.writeFile(out_file_commands,commands);
def parseInfoFile(out_file_text,lim=None):
lines=util.readLinesFromFile(out_file_text);
if lim is not None:
lines=lines[:lim];
h5_files=[];
img_files=[];
img_sizes=[];
for line_curr in lines:
str_split=line_curr.split(' ');
h5_files.append(str_split[0]);
img_files.append(str_split[1]);
img_sizes.append(tuple([int(i) for i in str_split[2:]]));
return h5_files,img_files,img_sizes
def writeClassTextFile(train_val_txt, path_to_im, out_file):
lines = util.readLinesFromFile(train_val_txt)
pos_im = []
lines_split = [line.split(' ', 1) for line in lines]
for idx, line_split in enumerate(lines_split):
num = int(line_split[1])
pos_im = [
line_split[0] for line_split in lines_split if int(line_split[1]) >= 0
]
ims = [
os.path.join(path_to_im, pos_im_curr + '.jpg')
for pos_im_curr in pos_im
]
util.writeFile(out_file, ims)
def script_saveFloAndVizFromTestFile(neg_file,out_dir,gpu=0):
pos_lines=util.readLinesFromFile(neg_file);
pos_img_files=[line[:line.index(' ')] for line in pos_lines];
print len(pos_img_files);
print pos_img_files[0];
gpu=1;
model_file='/home/maheenrashid/Downloads/debugging_jacob/optical_flow_prediction_test/examples/opticalflow/final.caffemodel'
clusters_file='/home/maheenrashid/Downloads/debugging_jacob/optical_flow_prediction_test/examples/opticalflow/clusters.mat';
dir_flo=os.path.join(out_dir,'flo');
util.mkdir(dir_flo);
dir_flo_viz=os.path.join(out_dir,'flo_viz');
util.mkdir(dir_flo_viz);
po.script_saveFlosAndViz(pos_img_files,dir_flo,dir_flo_viz,gpu,model_file,clusters_file,overwrite=True);
def us_getFilePres(gt_file, out_dir_us, post_us, num_iter, batch_us):
files_gt = []
files_pred = []
im_paths = util.readLinesFromFile(gt_file)
im_paths = [im_path[:im_path.index(' ')] for im_path in im_paths]
num_gt = len(im_paths)
count = 0
for batch_num in range(num_iter):
for im_num in range(batch_us):
file_pre = str(batch_num + 1) + '_' + str(im_num + 1)
file_gt = file_pre + post_us[0]
file_pred = file_pre + post_us[1]
files_gt.append(os.path.join(out_dir_us, file_gt))
files_pred.append(os.path.join(out_dir_us, file_pred))
files_gt = files_gt[:num_gt]
files_pred = files_pred[:num_gt]
return im_paths, files_gt, files_pred
def modifyHumanFile(orig_file, new_file):
data = util.readLinesFromFile(orig_file)
data = [
tuple([idx] + data_curr.split(' '))
for idx, data_curr in enumerate(data)
]
p = multiprocessing.Pool(multiprocessing.cpu_count())
new_lines = p.map(modifyHumanFileMultiProc, data)
# new_lines=[];
# for idx,(im_file,npy_file) in enumerate(data):
# print idx,len(data);
# im=scipy.misc.imread(im_file);
# im_size=im.shape;
# line_curr=npy_file+' '+str(im.shape[0])+' '+str(im.shape[1]);
# new_lines.append(line_curr);
print len(new_lines)
print new_lines[0]
util.writeFile(new_file, new_lines)
def writeMetaInfoToDb(path_to_db,out_files,idx_global,class_ids_all,path_to_data):
mani=Tube_Manipulator(path_to_db);
mani.openSession();
for out_file_idx,out_file in enumerate(out_files):
if out_file_idx%100==0:
print out_file_idx,len(out_files)
in_file_text=out_file.replace('.npz','.txt');
patch_files=util.readLinesFromFile(in_file_text);
# print out_file,in_file_text,len(patch_files);
for idx_img_file,img_file in enumerate(patch_files):
img_path=img_file;
img_path_split=img_path.split('/');
img_path_split=[segment for segment in img_path_split if segment!=''];
mat_name=img_path_split[-3];
class_id_pascal=mat_name[:mat_name.index('_')];
video_id=int(mat_name[mat_name.index('_')+1:mat_name.rindex('_')]);
shot_id=int(mat_name[mat_name.rindex('_')+1:]);
tube_id=int(img_path_split[-2]);
frame_id=img_path_split[-1];
frame_id=int(frame_id[:frame_id.index('.')]);
# frame_id+=1
class_idx_pascal=class_ids_all.index(class_id_pascal);
deep_features_path=out_file;
deep_features_idx=idx_img_file;
layer='fc7';
frame_path=getFramePath(path_to_data,class_id_pascal,video_id,shot_id,frame_id+1)
assert os.path.exists(frame_path);
mani.insert(idx_global, img_path, frame_id, video_id, tube_id, shot_id, frame_path=frame_path, layer=layer, deep_features_path=deep_features_path, deep_features_idx=deep_features_idx, class_id_pascal=class_id_pascal, class_idx_pascal=class_idx_pascal,commit=False);
idx_global+=1;
mani.session.commit();
mani.closeSession();
return idx_global;
def readGTFile(file_curr):
lines = util.readLinesFromFile(file_curr)
im_paths = []
ims_size = []
annos_all = []
for line_curr in lines:
line_split = line_curr.rsplit(None, 14)
im_paths.append(line_split[0])
im_size = line_split[1:1 + 4]
im_size = [int(num) for num in im_size]
im_size = [im_size[2] - im_size[0], im_size[3] - im_size[1]]
ims_size.append(im_size)
annos = line_split[1 + 4:]
annos = parseAnnoStr(annos)
annos_all.append(annos)
return im_paths, ims_size, annos_all
def script_writeHTMLStitchedFlos(out_file_html,out_file,out_dir,grid_sizes=[1,2,4,8],grid_dir_pre='grid_flo_viz_'):
img_paths=util.readLinesFromFile(out_file);
viz_dirs=[os.path.join(out_dir,grid_dir_pre+str(num)) for num in grid_sizes];
img_paths_html=[];
captions=[];
for img_path in img_paths:
img_name=img_path[img_path.rindex('/')+1:img_path.rindex('.')];
img_paths_html_curr=[util.getRelPath(img_path)];
captions_curr=['im']
for viz_dir in viz_dirs:
print viz_dir,img_path
img_path_curr=[os.path.join(viz_dir,file_curr) for file_curr in os.listdir(viz_dir) if file_curr.startswith(img_name)][0];
img_paths_html_curr.append(util.getRelPath(img_path_curr));
captions_curr.append(viz_dir[viz_dir.rindex('/')+1:]);
img_paths_html.append(img_paths_html_curr);
captions.append(captions_curr)
visualize.writeHTML(out_file_html,img_paths_html,captions);
def writeUniqueTrainingDataInfo(training_data_text,out_file_text):
lines=util.readLinesFromFile(training_data_text);
img_paths=[line[:line.index(' ')] for line in lines];
p=multiprocessing.Pool(multiprocessing.cpu_count());
vals=p.map(getDataSetAndVideoName,img_paths);
vals_uz=zip(*vals);
datasets=np.array(vals_uz[0]);
videos=np.array(vals_uz[1]);
new_tuples=[];
for dataset_curr in np.unique(datasets):
idx_rel=np.where(datasets==dataset_curr)[0];
videos_rel=videos[idx_rel];
videos_rel=np.unique(videos_rel);
for video_curr in videos_rel:
tuple_curr=(dataset_curr,video_curr)
new_tuples.append(tuple_curr);
vals_uni=[' '.join(val_curr) for val_curr in new_tuples];
util.writeFile(out_file_text,vals_uni);
def script_saveDetectedImages():
out_file = '/disk2/aprilExperiments/horses/list_of_frames.txt'
out_dir_dets = '/disk2/aprilExperiments/horses/frames_with_detections'
post_pend = '_horse_detections.npy'
post_pend_im = '_horse_detections.png'
class_name = 'horse'
thresh = 0.5
frames = util.readLinesFromFile(out_file)
# frames=['/disk2/aprilExperiments/horses/mediaFromPPT_frames/media7_00014.jpg']
for idx_frame_curr, frame_curr in enumerate(frames):
if idx_frame_curr % 100 == 0:
print idx_frame_curr, len(frames)
frame_name = frame_curr[frame_curr.rindex('/') + 1:]
frame_name = frame_name[:frame_name.rindex('.')]
video_name = frame_name[:frame_name.index('_')]
dets_file = os.path.join(out_dir_dets, video_name,
frame_name + post_pend)
out_file_im = dets_file.replace('.npy', '.png')
saveDets(frame_curr, class_name, dets_file, out_file_im, thresh)
def script_saveNonHumanOnlyNeg():
neg_file='/disk2/marchExperiments/deep_proposals/negatives.txt';
out_dir='/disk2/aprilExperiments/negatives_npy_onlyHuman'
lines=util.readLinesFromFile(neg_file);
npy_files=[line[line.index(' ')+1:] for line in lines];
npy_file_names=util.getFileNames(npy_files);
exists=0;
for idx_npy_file_name,npy_file_name in enumerate(npy_file_names):
if idx_npy_file_name%100==0:
print idx_npy_file_name;
file_curr=os.path.join(out_dir,npy_file_name);
if os.path.exists(file_curr):
exists+=1;
else:
zeros=np.zeros((0,4));
np.save(file_curr,zeros);
print exists,len(npy_file_names);
def saveBBoxImage(out_file, path_to_anno, out_dir_im, class_name='horse'):
files = util.readLinesFromFile(out_file)
just_names = util.getFileNames(files, ext=False)
annotations = [
os.path.join(path_to_anno, just_name + '.xml')
for just_name in just_names
]
print len(annotations)
for im_file, anno, just_name in zip(files, annotations, just_names):
out_file_pre = os.path.join(out_dir_im, just_name + '_')
obj = untangle.parse(anno)
for idx_object_curr, object_curr in enumerate(obj.annotation.object):
if object_curr.name.cdata == class_name:
out_file = out_file_pre + str(idx_object_curr) + '.jpg'
# if os.path.exists(out_file):
# continue;
bnd_box = [
object_curr.bndbox.xmin.cdata,
object_curr.bndbox.ymin.cdata,
object_curr.bndbox.xmax.cdata,
object_curr.bndbox.ymax.cdata
]
bnd_box = [int(coord) for coord in bnd_box]
# print bnd_box;
im = scipy.misc.imread(im_file)
if len(im.shape) < 3:
crop = im[bnd_box[1]:bnd_box[3], bnd_box[0]:bnd_box[2]]
else:
crop = im[bnd_box[1]:bnd_box[3], bnd_box[0]:bnd_box[2], :]
print out_file
scipy.misc.imsave(out_file, crop)
def saveFlosFromValFile(out_dir_model,val_file,num_to_pick,model_file,clusters_file,gpu,train_val_file=None,overwrite=False):
out_dir_flo=os.path.join(out_dir_model,'flo');
out_dir_flo_viz=os.path.join(out_dir_model,'flo_viz');
util.mkdir(out_dir_flo);util.mkdir(out_dir_flo_viz)
img_paths=util.readLinesFromFile(val_file);
img_paths=[img_path[:img_path.index(' ')] for img_path in img_paths];
class_names=[file_curr[:file_curr.index('_')] for file_curr in util.getFileNames(img_paths)];
classes=list(set(class_names));
class_names=np.array(class_names);
img_paths_test=[];
for class_curr in classes:
idx_rel=np.where(class_names==class_curr)[0];
idx_rel=idx_rel[:num_to_pick];
img_paths_test.extend([img_paths[idx_curr] for idx_curr in idx_rel]);
po.script_saveFlosAndViz(img_paths_test,out_dir_flo,out_dir_flo_viz,gpu,model_file,clusters_file,train_val_file=train_val_file,overwrite=overwrite);
return img_paths_test;
def main():
all_dirs_file = '/disk2/februaryExperiments/training_jacob/all_dirs.txt'
command_file_pre = '/disk2/februaryExperiments/training_jacob/commands_training_data_'
train_data_file = '/disk2/februaryExperiments/training_jacob/caffe_files/train.txt'
check_file = 'done.mat'
num_proc = 12
# command_files = script_writeCommandsForPreprocessing(all_dirs_file,command_file_pre,num_proc,check_file);
all_dirs = util.readLinesFromFile(all_dirs_file)
# all_dirs=all_dirs[:10];
random.shuffle(all_dirs)
strings = []
for no_dir_curr, dir_curr in enumerate(all_dirs):
print no_dir_curr, dir_curr
dir_curr = dir_curr[:-1]
curr_flos = [
os.path.join(dir_curr, curr_flo)
for curr_flo in os.listdir(dir_curr) if curr_flo.endswith('.tif')
]
for curr_flo in curr_flos:
curr_im = curr_flo.replace('.tif', '.jpg')
assert os.path.exists(curr_im)
string_curr = curr_im + ' ' + curr_flo + ' '
strings.append(string_curr)
print len(strings)
# print strings[:3];
# random.shuffle(strings);
util.writeFile(train_data_file, strings)
# with open (train_data_file,'wb') as f:
# for im_curr,flo_curr in zip(ims,flos):
# string_curr=im_curr+' '+flo_curr+'\n';
# f.write(string_curr);
return
dirs = getRemainingDirs(util.readLinesFromFile(all_dirs_file), check_file)
last_lines = []
for dir_curr in dirs:
last_lines.append(
util.readLinesFromFile(os.path.join(dir_curr, 'log.txt'))[-2])
print set(last_lines)
return
meta_dirs_image = [
'/disk2/image_data_moved',
'/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/image_data'
]
meta_dirs_flo = [
'/disk2/flow_data',
'/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/flow_data'
]
sub_dirs_file = 'all_sub_dirs.txt'
out_dir = '/disk2/februaryExperiments/training_jacob'
out_file_correspondences = os.path.join(out_dir,
'im_flo_correspondences.p')
proto_file = 'deploy.prototxt'
out_file = os.path.join(out_dir, 'im_flo_files.p')
out_dir = '/disk2/februaryExperiments/training_jacob/training_data'
mat_file = 'im_flo_files.mat'
if not os.path.exists(out_dir):
os.mkdir(out_dir)
im_flo_dirs = pickle.load(open(out_file_correspondences, 'rb'))
[im_dirs, flo_dirs] = zip(*im_flo_dirs)
for im_dir, flo_dir in im_flo_dirs:
script_saveMatFiles(flo_dir, im_dir, out_dir, mat_file, proto_file)
def script_saveMatFiles(flo_dir, im_dir, out_dir, mat_file, proto_file):
#get video name
video_name = flo_dir[:-1]
video_name = video_name[video_name.rindex('/') + 1:]
print video_name
#get flo files
flo_files = [
os.path.join(flo_dir, file_curr) for file_curr in os.listdir(flo_dir)
if file_curr.endswith('.flo')
]
flo_files.sort()
#get im files
im_files = util.readLinesFromFile(os.path.join(flo_dir, 'im_1.txt'))
old_dir = im_files[0][:im_files[0].rindex('/') + 1]
#if dirs have changed, replace the paths
if im_dir != old_dir:
im_files = [im_curr.replace(old_dir, im_dir) for im_curr in im_files]
#get batch size
batch_size = getBatchSizeFromDeploy(os.path.join(flo_dir, proto_file))
#get batch info
batch_num = [
int(file_curr[file_curr.rindex('-') + 1:file_curr.rindex('(')])
for file_curr in flo_files
]
batch_num = np.array(batch_num)
batch_ids = list(set(batch_num))
batch_ids.sort()
flo_files_all = []
im_files_all = []
for batch_no in batch_ids:
idx_rel = np.where(batch_num == batch_no)[0]
flo_files_curr = []
im_files_curr = []
for idx_curr in idx_rel:
flo_file = flo_files[idx_curr]
im_no = int(flo_file[flo_file.rindex('(') +
1:flo_file.rindex(')')])
im_corr = im_files[batch_no * batch_size + im_no]
flo_files_curr.append(flo_file)
im_files_curr.append(im_corr)
flo_files_all.append(flo_files_curr)
im_files_all.append(im_files_curr)
#save as mat with flofiles, im_files, and out_dir;
for idx_batch_no, batch_no in enumerate(batch_ids):
flo_files = flo_files_all[idx_batch_no]
im_files = im_files_all[idx_batch_no]
out_dir_mat = os.path.join(out_dir, video_name + '_' + str(batch_no))
# print out_dir_mat
if not os.path.exists(out_dir_mat):
os.mkdir(out_dir_mat)
out_file = os.path.join(out_dir_mat, mat_file)
print out_file
mat_data = {'flo_files': flo_files, 'im_files': im_files}
scipy.io.savemat(out_file, mat_data)
import util
import visualize
import os
import preprocessing_data
import scipy.misc
import cv2
# from IPython.display import Image
val_list_file = '/home/laoreja/new-deep-landmark/dataset/train/aflw_valImageList.txt'
out_dir = '/home/SSD3/maheen-data/temp'
dir_server = '/home/SSD3/maheen-data'
val_list = util.readLinesFromFile(val_list_file)
path_im, bbox, anno_points = preprocessing_data.parseAnnoFile(val_list_file)
print len(path_im), len(bbox), len(anno_points)
print path_im[0], bbox[0], anno_points[0]
print len(val_list)
# plt.ion();
for idx, (im_path, bbox_curr,
anno_points_curr) in enumerate(zip(path_im, bbox, anno_points)):
print im_path
out_file_curr = os.path.join(out_dir,
str(idx) + '.jpg')
im = cv2.imread(im_path)
cv2.rectangle(im, (bbox_curr[0], bbox_curr[2]),
(bbox_curr[1], bbox_curr[3]), (255, 0, 0))
def experiment_nnPatches(params):
out_file_text = params.out_file_text
class_ids = params.class_id
img_paths = util.readLinesFromFile(out_file_text)
class_id_idx_tuples = []
for img_path in img_paths:
class_id = img_path[:img_path.rindex('_')]
class_id = class_id[class_id.rindex('_') + 1:]
class_idx = class_ids.index(class_id)
class_idx = params.class_idx[class_idx]
class_id_idx_tuples.append((class_id, class_idx))
file_names_mat, object_indices = recreateOriginalPaths(
params.path_to_annotation, img_paths, returnObject_idx=True)
print 'getting azimuths'
azimuths = [
getObjectStruct(file_name, object_idx).viewpoint.azimuth_coarse
for file_name, object_idx in zip(file_names_mat, object_indices)
]
if params.out_file_layers is None:
print 'running layers part'
out_file_layers = caffe_wrapper.saveFeaturesOfLayers(
out_file_text,
params.path_to_classify,
params.gpu_no,
params.layers,
ext='jpg',
out_file=params.out_file_pre,
meanFile=params.caffe_mean,
deployFile=params.caffe_deploy,
modelFile=params.caffe_model,
images_dim=params.images_dim)
params = params._replace(out_file_layers=out_file_layers)
out_file_layers = params.out_file_layers
print 'writing to db'
for layer in params.layers:
vals = np.load(out_file_layers)
indices, distances = nearest_neighbor.doCosineDistanceNN(
vals[layer], numberOfN=None)
mani = Pascal3D_Manipulator(params.db_path_out)
mani.openSession()
for idx in range(len(img_paths)):
mani.insert(idx,
img_paths[idx],
layer,
out_file_layers,
class_id_idx_tuples[idx][0],
class_id_idx_tuples[idx][1],
params.caffe_model,
azimuth=azimuths[idx],
neighbor_index=indices[idx],
neighbor_distance=distances[idx],
trainedClass=params.trainFlag,
commitFlag=False)
mani.closeSession()
return params
def main():
# dir_meta='/disk2/flow_data';
# dir_meta_old='/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/flow_data';
# deploy_file='deploy.prototxt';
# dir_mids=[os.path.join(dir_meta,dir_mid) for dir_mid in os.listdir(dir_meta) if os.path.isdir(os.path.join(dir_meta,dir_mid))]
# dirs_left=[os.path.join(dir_mid,dir_curr) for dir_mid in dir_mids for dir_curr in os.listdir(dir_mid) if os.path.isdir(os.path.join(dir_mid,dir_curr))]
# dirs_left=[os.path.join(dir_mid,dir_curr) for dir_mid in dirs_left for dir_curr in os.listdir(dir_mid) if os.path.isdir(os.path.join(dir_mid,dir_curr))]
# print len(dirs_left);
# print dirs_left[0];
# for dir_curr in dirs_left:
# deploy_curr=os.path.join(dir_curr,deploy_file);
# print deploy_curr
# data=[];
# with open(deploy_curr,'r') as f:
# data = f.read()
# with open(deploy_curr+'_backup','w') as f:
# f.write(data);
# data = data.replace(dir_meta_old, dir_meta)
# with open(deploy_curr, "w") as f:
# f.write(data);
# return
# video_list_file='/disk2/video_data/video_list.txt'
# path_to_video_meta='/disk2/video_data';
# path_to_flo_meta='/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/flow_data';
# path_to_im_meta='/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/image_data';
# video_files=util.readLinesFromFile(video_list_file);
# # image_dirs=[dir_curr.replace(path_to_video_meta,path_to_im_meta)[:-4] for dir_curr in video_files];
# # flo_dirs=[dir_curr.replace(path_to_video_meta,path_to_flo_meta)[:-4] for dir_curr in video_files];
# flo_dirs=pickle.load(open('/disk2/temp/dirs_done.p','rb'));
# image_dirs=[dir_curr.replace(path_to_flo_meta,path_to_im_meta) for dir_curr in flo_dirs];
# print len(image_dirs)
# out_dir='/disk2/image_data_moved';
# out_file='/disk2/image_data_moved/mv_commands_2.txt'
# commands=[];
# image_dirs_to_move=image_dirs[5000:7000];
# for image_dir in image_dirs_to_move:
# image_dir=util.escapeString(image_dir);
# new_dir=image_dir.replace(path_to_im_meta,out_dir);
# command='mkdir -p '+new_dir+';';
# command=command+'mv '+image_dir+'/* '+new_dir;
# commands.append(command);
# util.writeFile('/disk2/image_data_moved/dirs_moved_2.txt',image_dirs_to_move);
# util.writeFile(out_file,commands);
# return
video_list_file = '/disk2/video_data/video_list.txt'
path_to_video_meta = '/disk2/video_data'
# path_to_flo_meta='/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/flow_data';
path_to_flo_meta = '/disk2/flow_data'
path_to_im_meta = '/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/image_data'
video_files = util.readLinesFromFile(video_list_file)
# image_dirs=[dir_curr.replace(path_to_video_meta,path_to_im_meta)[:-4] for dir_curr in video_files];
# flo_dirs=[dir_curr.replace(path_to_video_meta,path_to_flo_meta)[:-4] for dir_curr in video_files];
flo_dirs = pickle.load(open('/disk2/temp/dirs_done_disk2.p', 'rb'))
image_dirs = [
dir_curr.replace(path_to_flo_meta, path_to_im_meta)
for dir_curr in flo_dirs
]
print len(image_dirs)
finished = []
i = 0
for image_dir, flo_dir in zip(image_dirs, flo_dirs):
print i
count_im_command = 'ls ' + os.path.join(util.escapeString(image_dir),
'*.ppm') + '| wc -l'
count_flo_command = 'ls ' + os.path.join(util.escapeString(flo_dir),
'*.flo') + '| wc -l'
# im_count=int(subprocess.check_output(count_im_command,shell=True));
# flo_count=int(subprocess.check_output(count_flo_command,shell=True));
im_count = len([
file_curr for file_curr in os.listdir(image_dir)
if file_curr.endswith('.ppm')
])
flo_count = len([
file_curr for file_curr in os.listdir(flo_dir)
if file_curr.endswith('.flo')
])
print i, flo_count, im_count
if flo_count + 1 == im_count:
finished.append(1)
else:
finished.append(0)
i += 1
finished = np.array(finished)
print 'done', sum(finished == 1)
print 'not done', sum(finished == 0)
pickle.dump([finished, image_dirs], open('/disk2/temp/to_rerun.p', 'wb'))
return
dir_flownet_meta = '/home/maheenrashid/Downloads/flownet/flownet-release/models/flownet'
caffe_bin = os.path.join(dir_flownet_meta, 'bin/caffe')
path_to_model = os.path.join(dir_flownet_meta,
'model/flownet_official.caffemodel')
video_list_file = '/disk2/video_data/video_list.txt'
path_to_video_meta = '/disk2/video_data'
in_dir_meta = '/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/flow_data'
in_dir_meta = '/disk2/flow_data'
# if not os.path.exists(new_in_dir_meta):
# os.mkdir(new_in_dir_meta);
deploy_name = 'deploy.prototxt'
gpu = 0
dirs = [
dir_curr.replace(path_to_video_meta, in_dir_meta)[:-4]
for dir_curr in util.readLinesFromFile(video_list_file)
]
dirs = [dir_curr for dir_curr in dirs if os.path.exists(dir_curr)]
counts = [
len(os.listdir(dir_curr)) for dir_curr in dirs
if os.path.exists(dir_curr)
]
dirs_left = []
dirs_done = []
for idx_count, count in enumerate(counts):
if count == 4:
dirs_left.append(dirs[idx_count])
# dir_curr=dirs[idx_count]
# deploy_curr=os.path.join(dir_curr,deploy_name);
# im_file=os.path.join(dir_curr,'im_1.txt');
# batch_size = sum(1 for line in open(im_file))
# old_str='batch_size: '+str(int(ceil(batch_size/5)));
# print old_str,
# batch_size = int(ceil(batch_size/8));
# new_str='batch_size: '+str(batch_size);
# print new_str
# data=[];
# with open(deploy_curr,'r') as f:
# data = f.read()
# # print data[:300];
# assert old_str in data;
# data = data.replace(old_str, new_str)
# # print data[:300];
# with open(deploy_curr, "w") as f:
# f.write(data);
# out_dir_curr=dir_curr.replace(in_dir_meta,new_in_dir_meta);
#mkdir of new location
# mkdir_command='mkdir -p '+util.escapeString(out_dir_curr)
# print mkdir_command
# subprocess.call(mkdir_command, shell=True)
#mv contents from old to new
# mv_command='mv '+util.escapeString(dir_curr)+'/* '+util.escapeString(out_dir_curr);
# print mv_command
# subprocess.call(mv_command, shell=True)
#append new to dirs_left
# dirs_left.append(out_dir_curr);
# raw_input();
else:
dirs_done.append(dirs[idx_count])
print min(counts)
counts = np.array(counts)
print sum(counts == 4)
print len(dirs_left)
mid_point = len(dirs_left) / 2
print mid_point, len(dirs_left) - mid_point
out_file_commands = '/disk2/januaryExperiments/gettingFlows/flownet_commands_left_0.txt'
gpu = 0
# writeCommands_hacky(out_file_commands,dirs_left[:mid_point],caffe_bin,deploy_name,path_to_model,gpu)
out_file_commands = '/disk2/januaryExperiments/gettingFlows/flownet_commands_left_1.txt'
gpu = 1
# writeCommands_hacky(out_file_commands,dirs_left[mid_point:],caffe_bin,deploy_name,path_to_model,gpu)
print len(dirs_done)
pickle.dump(dirs_done, open('/disk2/temp/dirs_done_disk2.p', 'wb'))
return
video_list_file = '/disk2/video_data/video_list.txt'
path_to_video_meta = '/disk2/video_data'
in_dir_meta = '/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/image_data'
out_dir_meta = '/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/flow_data'
path_to_deploy = '/disk2/januaryExperiments/gettingFlows/deploy_template.prototxt'
out_file_commands = '/disk2/januaryExperiments/gettingFlows/flownet_commands.txt'
dir_flownet_meta = '/home/maheenrashid/Downloads/flownet/flownet-release/models/flownet'
path_to_sizer = os.path.join(dir_flownet_meta, 'bin/get_image_size')
caffe_bin = os.path.join(dir_flownet_meta, 'bin/caffe')
path_to_model = os.path.join(dir_flownet_meta,
'model/flownet_official.caffemodel')
text_1 = 'im_1.txt'
text_2 = 'im_2.txt'
deploy_file = 'deploy.prototxt'
gpu = 0
params_dict = {}
params_dict['video_list_file'] = video_list_file
params_dict['path_to_video_meta'] = path_to_video_meta
params_dict['in_dir_meta'] = in_dir_meta
params_dict['out_dir_meta'] = out_dir_meta
params_dict['path_to_deploy'] = path_to_deploy
params_dict['out_file_commands'] = out_file_commands
params_dict['dir_flownet_meta'] = dir_flownet_meta
params_dict['path_to_sizer'] = path_to_sizer
params_dict['caffe_bin'] = caffe_bin
params_dict['path_to_model'] = path_to_model
params_dict['text_1'] = text_1
params_dict['text_2'] = text_2
params_dict['deploy_file'] = deploy_file
params_dict['gpu'] = gpu
params = createParams('writeFlownetCommands')
params = params(**params_dict)
# script_writeFlownetCommands(params);
commands = util.readLinesFromFile(params.out_file_commands)
commands = [c.replace('-gpu 1', '-gpu 0') for c in commands]
util.writeFile(params.out_file_commands, commands)
pickle.dump(params._asdict(),
open(params.out_file_commands + '_meta_experiment.p', 'wb'))
return
video_list_file = '/disk2/video_data/video_list.txt'
path_to_im_meta = '/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/image_data'
path_to_video_meta = '/disk2/video_data'
commands_file_text = '/disk2/januaryExperiments/gettingFlows/resize_commands.txt'
video_list = util.readLinesFromFile(video_list_file)
print len(video_list)
image_dirs = [
video_curr.replace(path_to_video_meta, path_to_im_meta)[:-4]
for video_curr in video_list
]
print len(image_dirs), image_dirs[0]
image_dirs = image_dirs[:1]
commands = []
command_conv = ['convert', '-resize 512x384']
for image_dir in image_dirs:
image_list = [
os.path.join(image_dir, im) for im in os.listdir(image_dir)
if im.endswith('.ppm')
]
for image_curr in image_list:
command_curr = [
command_conv[0], image_curr, command_conv[1], image_curr
]
command_curr = ' '.join(command_curr)
commands.append(command_curr)
print len(commands)
print commands[0]
util.writeFile(commands_file_text, commands)
return
video_list_file = '/disk2/video_data/video_list.txt'
path_to_im_meta = '/media/maheenrashid/e5507fe3-2bff-4cbe-bc63-400de6deba92/maheen_data/image_data'
path_to_video_meta = '/disk2/video_data'
path_to_txt_1 = '/disk2/januaryExperiments/gettingFlows/temp_im_1.txt'
path_to_txt_2 = '/disk2/januaryExperiments/gettingFlows/temp_im_2.txt'
video_list = util.readLinesFromFile(video_list_file)
print len(video_list)
image_dirs = [
video_curr.replace(path_to_video_meta, path_to_im_meta)[:-4]
for video_curr in video_list
]
print len(image_dirs), image_dirs[0]
list_1 = []
list_2 = []
for image_dir in image_dirs[:10]:
list_1_curr, list_2_curr = getImageListForFlow(image_dir)
list_1.extend(list_1_curr[:3])
list_2.extend(list_2_curr[:3])
assert len(list_1) == len(list_2)
util.writeFile(path_to_txt_1, list_1)
util.writeFile(path_to_txt_2, list_2)