-
Notifications
You must be signed in to change notification settings - Fork 0
/
flowEval.py
483 lines (366 loc) · 15.7 KB
/
flowEval.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
import numpy as np;
import os;
import util;
import multiprocessing;
import random;
import math;
import processOutput as po;
import script_testJacob as stj;
import subprocess;
import cv2;
def getEPE(flo_gt,flo_pred):
sqr=np.power(flo_gt-flo_pred,2);
# print sqr.shape
summed=sqr[:,:,0]+sqr[:,:,1];
# print summed.shape
sqr_rt=np.power(summed,0.5);
epe=np.mean(sqr_rt);
return epe,sqr_rt
def getOrientationSimilarity(flo_gt,flo_pred):
flo_gt=np.vstack((flo_gt[:,:,0].ravel(),flo_gt[:,:,1].ravel()));
flo_pred=np.vstack((flo_pred[:,:,0].ravel(),flo_pred[:,:,1].ravel()));
flo_gt_norms=np.power(np.sum(np.power(flo_gt,2),0),0.5);
flo_pred_norms=np.power(np.sum(np.power(flo_pred,2),0),0.5);
dots=[];
for val_idx in range(flo_gt.shape[1]):
dots.append(np.dot(flo_gt[:,val_idx],flo_pred[:,val_idx]));
dots=np.array(dots);
dots=np.abs(dots);
printflag=False;
if np.sum(flo_gt_norms==0)>0 or np.sum(flo_pred_norms==0)>0:
print 'zero_norms_orientation'
print np.sum(flo_gt_norms==0),np.sum(flo_pred_norms==0);
ors_old=dots/(flo_gt_norms*flo_pred_norms);
flo_gt_norms[flo_gt_norms==0]=np.finfo(float).eps;
flo_pred_norms[flo_pred_norms==0]=np.finfo(float).eps;
printflag=True;
ors=dots/(flo_gt_norms*flo_pred_norms);
# ors=dots/flo_pred_norms;
if printflag:
print np.mean(ors),np.mean(ors_old);
return np.mean(ors),ors;
def getDirectionSimilarity(flo_gt,flo_pred):
flo_gt=np.vstack((flo_gt[:,:,0].ravel(),flo_gt[:,:,1].ravel()));
flo_pred=np.vstack((flo_pred[:,:,0].ravel(),flo_pred[:,:,1].ravel()));
flo_gt_norms=np.power(np.sum(np.power(flo_gt,2),0),0.5);
flo_pred_norms=np.power(np.sum(np.power(flo_pred,2),0),0.5);
dots=[];
for val_idx in range(flo_gt.shape[1]):
dots.append(np.dot(flo_gt[:,val_idx],flo_pred[:,val_idx]));
dots=np.array(dots);
printflag=False;
if np.sum(flo_gt_norms==0)>0 or np.sum(flo_pred_norms==0)>0:
print 'zero_norms_direction'
print np.sum(flo_gt_norms==0),np.sum(flo_pred_norms==0);
ors_old=dots/(flo_gt_norms*flo_pred_norms);
flo_gt_norms[flo_gt_norms==0]=np.finfo(float).eps;
flo_pred_norms[flo_pred_norms==0]=np.finfo(float).eps;
printflag=True;
ors=dots/(flo_gt_norms*flo_pred_norms);
# ors=dots/flo_pred_norms;
if printflag:
print np.mean(ors),np.mean(ors_old);
return np.mean(ors),ors;
def getEPEWrapper((path_gt,path_pred,idx,includeBigMat,error_type)):
print idx;
flo_gt=np.load(path_gt);
flo_pred=np.load(path_pred);
if error_type=='epe':
epe,epe_pix=getEPE(flo_gt,flo_pred);
elif error_type=='direction':
epe,epe_pix=getDirectionSimilarity(flo_gt,flo_pred);
elif error_type=='orientation':
epe,epe_pix=getOrientationSimilarity(flo_gt,flo_pred);
if includeBigMat:
ret_value=(epe,epe_pix)
else:
ret_value=epe;
return ret_value;
def getAllErrorsWrapper((path_gt,path_pred,idx,isFlo)):
print idx;
if isFlo:
flo_gt=util.readFlowFile(path_gt);
flo_pred=util.readFlowFile(path_pred);
else:
flo_gt=np.load(path_gt);
flo_pred=np.load(path_pred);
# print flo_gt.shape,flo_pred.shape,path_gt,path_pred;
# if flo_gt.shape!=flo_pred.shape:
# util.writeFlowFile(flo_pred,'/disk2/temp/flo_pred_bef.flo');
shape_old=flo_pred.shape;
flo_pred=cv2.resize(flo_pred,(flo_gt.shape[1],flo_gt.shape[0]));
flo_pred[:,:,0]=flo_pred[:,:,0]*float(flo_gt.shape[1])/shape_old[1];
flo_pred[:,:,1]=flo_pred[:,:,1]*float(flo_gt.shape[0])/shape_old[0];
# util.writeFlowFile(flo_pred,'/disk2/temp/flo_pred_aft.flo');
# util.writeFlowFile(flo_gt,'/disk2/temp/flo_gt.flo');
epe,_=getEPE(flo_gt,flo_pred);
dir_sim,_=getDirectionSimilarity(flo_gt,flo_pred);
or_sim,_=getOrientationSimilarity(flo_gt,flo_pred);
ret_value=(epe,dir_sim,or_sim)
return ret_value;
def getErrorMultiProc(dir_gt,dir_pred,np_files,out_file_err,isFlo=False):
args=[];
for idx_np_file,np_file in enumerate(np_files):
args.append((os.path.join(dir_gt,np_file),os.path.join(dir_pred,np_file),idx_np_file,isFlo));
print len(args);
print args[0];
# if error_type=='epe':
# list_errors=[];
# for arg_curr in args:
# list_errors.append(getAllErrorsWrapper(arg_curr));
p=multiprocessing.Pool(multiprocessing.cpu_count());
list_errors=p.map(getAllErrorsWrapper,args);
# for arg in args:
# getAllErrorsWrapper(arg);
list_errors=np.array(list_errors);
files=np.array(np_files);
# print list_errors.shape,files.shape
# print list_errors[:10],files[:10]
np.savez(out_file_err, files=files, errors=list_errors);
def splitFilesByClass(np_files):
class_dict={}
for file_curr in np_files:
class_curr=file_curr[:file_curr.index('_')];
if class_curr in class_dict:
class_dict[class_curr].append(file_curr);
else:
class_dict[class_curr]=[file_curr];
return class_dict;
def script_compYoutubePerf(dir_org,dir_new):
# list_files=[file_curr for file_curr in os.listdir(dir_org) if file_curr.endswith('.npz')];
# for list_file in list_files:
# arrs=np.load(os.path.join(dir_org,list_file))
arrs=np.load(dir_org)
errs_old=arrs['errors'];
list_files=arrs['files'];
classes_new=np.array([list_file[:list_file.index('_')] for list_file in list_files]);
classes_uni=np.unique(classes_new);
print list_files[0];
# print errs_old.shape;
# arrs=np.load(os.path.join(dir_new,list_file))
arrs=np.load(dir_new)
errs_new=arrs['errors'];
list_files=arrs['files'];
classes_old=np.array([list_file[:list_file.index('_')] for list_file in list_files]);
# print errs_new.shape;
# class_name=list_file[list_file.rindex('/')+1:]
# class_name=list_file[:list_file.index('_')];
# print class_name
classes_uni=np.unique(classes_new);
print classes_uni;
for class_curr in classes_uni:
print class_curr
print 'old';
print np.mean(errs_old[classes_old==class_curr,:],axis=0);
print 'new';
print np.mean(errs_new[classes_new==class_curr,:],axis=0);
print '___';
# break;
def getDataSetAndVideoName(line):
line_split=line.split('/');
# print line_split
data_set=line_split[3];
video=line_split[-1];
video=video[:video.index('.')];
return data_set,video
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 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 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 main():
training_data_text='/disk2/marchExperiments/finetuning_youtube_hmdb_llr/train_both.txt';
dir_meta='/disk2/mayExperiments';
# dir_meta='/group/leegrp/maheen_data'
dir_curr='flow_eval';
dir_train=os.path.join(dir_meta,'finetuning_youtube_hmdb_llr');
util.mkdir(dir_train);
train_new_text=os.path.join(dir_train,'train.txt');
out_file_eq=os.path.join(dir_train,'train_eq.txt');
out_file_idx=os.path.join(dir_train,'train_eq_idx.npy');
val_file_eq=os.path.join(dir_train,'val_eq.txt');
val_file_idx=os.path.join(dir_train,'val_eq_idx.npy');
val_new_text=os.path.join(dir_train,'val.txt');
percent_exclude=0.1;
out_file_text_org=os.path.join(dir_train,'train_info_org.txt');
out_file_text=os.path.join(dir_meta,dir_curr,'train_info.txt');
dir_test_meta=os.path.join(dir_meta,'test_youtube_flow');
util.mkdir(dir_test_meta);
# dir_test=os.path.join(dir_test_meta,'original_model');
# util.mkdir(dir_test);
# model_file='/home/maheenrashid/Downloads/debugging_jacob/optical_flow_prediction_test/examples/opticalflow/final.caffemodel'
dir_test=os.path.join(dir_test_meta,'50000_model');
util.mkdir(dir_test);
model_file='/disk2/mayExperiments/finetuning_youtube_hmdb_llr/OptFlow_youtube_hmdb_iter_50000.caffemodel'
clusters_file='/home/maheenrashid/Downloads/debugging_jacob/optical_flow_prediction_test/examples/opticalflow/clusters.mat';
gpu=0;
lines=util.readLinesFromFile(val_new_text);
img_paths=[line[:line.index(' ')] for line in lines];
# po.script_saveFlos(img_paths,dir_test,gpu,model_file,clusters_file,overwrite=False)
dir_pred=os.path.join(dir_test,'flo_files');
print dir_pred;
dir_gt=os.path.join(dir_test_meta,'gt_flow');
flo_files=[img_name+'.flo' for img_name in util.getFileNames(img_paths,False)];
# flo_files=flo_files[:1];
out_file_err=dir_test+'_err_all.npz';
# getErrorMultiProc(dir_gt,dir_pred,flo_files,out_file_err,isFlo=True)
out_file_err_old=os.path.join(dir_test_meta,'original_model_err_all.npy.npz');
script_compYoutubePerf(out_file_err_old,out_file_err)
return
writeTrainValTxtExcluded(train_new_text,val_new_text,out_file_text_org,training_data_text,percent_exclude)
writeUniqueTrainingDataInfo(train_new_text,out_file_text)
lines=util.readLinesFromFile(out_file_text_org);
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];
else:
if dataset in class_rec:
class_rec[dataset].append(video);
else:
class_rec[dataset]=[video];
for class_curr in class_rec.keys():
print class_curr,len(class_rec[class_curr])
return
paths_replace=['/disk2/marchExperiments','/group/leegrp/maheen_data'];
transfer_file=os.path.join(dir_test,'transfer.txt');
img_paths_replace=[file_curr.replace(paths_replace[0],paths_replace[1]).replace('.jpg','.flo').replace('images_transfer','images') for file_curr in img_paths];
print len(img_paths_replace);
print img_paths_replace[0];
util.writeFile(transfer_file,img_paths_replace);
print transfer_file;
return
writeUniqueTrainingDataInfo(training_data_text,out_file_text_org)
return
return
# dir_youtube=os.path.join(dir_meta,'youtube');
# dir_hmdb=os.path.join(dir_meta,'hmdb');
youtube_data=[dir_curr for dir_curr in os.listdir(dir_youtube)];
hmdb_data=[dir_curr for dir_curr in os.listdir(dir_hmdb)];
print len(youtube_data);
train_info=util.readLinesFromFile(out_file_text);
youtube_data_train=[];
hmdb_data_train=[];
for info_curr in train_info:
dataset=info_curr[:info_curr.index(' ')];
video=info_curr[info_curr.index(' ')+1:];
if dataset=='youtube':
youtube_data_train.append(video);
else:
hmdb_data_train.append(video);
youtube_data_train=set(youtube_data_train);
leftover_youtube=set(youtube_data).difference(youtube_data_train);
leftover_hmdb=set(hmdb_data).difference(hmdb_data_train);
flo_files_all=[];
for idx_hmdb,hmdb_curr in enumerate(leftover_hmdb):
print idx_hmdb,len(leftover_hmdb),hmdb_curr;
dir_curr=os.path.join(dir_hmdb,hmdb_curr,'images');
flo_files=[os.path.join(dir_curr,file_curr) for file_curr in os.listdir(dir_curr) if file_curr.endswith('.flo')];
if len(flo_files)==0:
print 'PROBLEM';
flo_files_all=flo_files_all+flo_files;
print len(flo_files_all);
print flo_files_all[0];
print flo_files_all[100];
util.writeFile(heldout_hmdb,flo_files_all);
# /group/leegrp/maheen_data/hmdb/AmericanGangster_drink_u_nm_np1_fr_med_40
# print leftover;
# print len(leftover_youtube);
# print len(leftover_hmdb),len(hmdb_data_train);
# out_dir='/disk2/mayExperiments/flow_eval'
# out_dir='/disk2/mayExperiments/flow_eval'
# out_file_text=os.path.join(out_dir,'train_info.txt');
if __name__=='__main__':
main();