forked from Dong-JinKim/DRCaptioning
-
Notifications
You must be signed in to change notification settings - Fork 0
/
model_name.txt
834 lines (608 loc) · 31.6 KB
/
model_name.txt
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
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
checkpoint1e5.t7
=>learning rate 1e-5 (batch 128)
checkpoint1e6.t7
=>learning rate 1e-6, fineturned from checkpoint1e5.t7 (batch 128)
checkpoint1e5_2.t7
=>batch size 64
checkpoint1e6_2.t7
=>batch size 128 , finetune from checkpoint trained.t7
checkpoint_base.t7
=>without spatial mask, learning rate 1e-6, finetuned from trained.t7
-
(LanguageModel - LocalizationLayer - Pairs)
checkpoint1e5_VRD.t7
=>learning rate 1e-5 (batch 128) on VRD dataset
checkpoint1e6_VRD.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
checkpoint1e6_2_VRD.t7
=>learning rate 1e-6 (batch 128) on VRD dataset fineturned from checkpoint1e5_VRD.t7
(LanguageModel2 - LocalizationLayer - Pairs2)
checkpoint1e6_VRD2.t7 (with BN)
=>learning rate 1e-6 (batch 128) on VRD dataset
didnt use spatial mask, use naive coordinate information
=>higher loss value than VRD1--------------------------------fail!!! don't use BN "left left left left ...."
checkpoint1e5_VRD2.t7 (with BN)
=>explode too soon. failed "left left left left ...."
(LanguageModel2 - LocalizationLayer - Pairs2)
checkpoint1e6_VRD2_noBN.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use naive coordinate information
no using BN
=> alway starts with "person"!!!!
--------------------------------------------------------------------
(LanguageModel_union(512,S=0) - LocalizationLayer_union1(pairs_RPN) )
checkpoint1e6_VRD2_union.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "only union BB" as input
(LanguageModel_union(512*3,S=0)- LocalizationLayer_union2(pairs_RPN))
checkpoint1e6_VRD2_union2.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat+subj+obj" feat as input
(LanguageModel_union(512*3,S=0)- LocalizationLayer_union2(pairs_RPN) )
checkpoint_1e6_VRD2_union2dropout.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat+subj+obj" feat as input (w/ dropout)
(LanguageModel_union(512*3,S=0)- LocalizationLayer_union2(pairs_RPN) )
checkpoint_1e6_VRD2_union2thin.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat+subj+obj" feat as input (1536*256+256*512 instead of 1536*512)
(LanguageModel_union(512*3,S=0)- LocalizationLayer_union2(pairs_RPN) )
checkpoint_1e6_VRD2_union2thin2_RE.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat+subj+obj" feat as input (1536*128+128*512 instead of 1536*512)
(LanguageModel_union(512*3,S=0)- LocalizationLayer_union2(pairs_RPN) )
checkpoint_1e6_VRD2_union2thin3.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat+subj+obj" feat as input (1536*256+256*512 instead of 1536*512)
switch order of 1*1covn and ave pooling
----------------------------------------------------------------------
---------------------------------------------------------------------
(LanguageModel_union(512*3,S=64) - LocalizationLayer_union2(pairs_RPN1) )
checkpoint_VRD2_union_S1_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat+subj+obj"+ "spaitial mask" information
(LanguageModel_union(512*3,S=10) - LocalizationLayer_union2(pairs_RPN2))
(Deleted!!!!)checkpoint_VRD2_union_S2_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat+subj+obj"+ "coordinate" information
---------------------------------------------------------------------
--------------------------------------------------------
(LanguageModel_union(512,S=64) - LocalizationLayer_union1(pairs_RPN1))
checkpoint_VRD2_union2_S1_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat"+ "spaitial mask" information
(LanguageModel_union(512,S=10) - LocalizationLayer_union1(pairs_RPN2) )
checkpoint_VRD2_union2_S2_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat"+ "coordinate" information
(LanguageModel_union(512,S=10) - LocalizationLayer_union1(pairs_RPN2))
checkpoint_VRD2_union2_S2_1e6_test.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat"+ "coordinate" information
use [0000] as spatial feat
(LanguageModel_union(512,S=10) - LocalizationLayer_union1(pairs_RPN2))
checkpoint_VRD2_union2_S2dropout_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat"+ "coordinate" information
added dropout
(LanguageModel_union(512,S=6) - LocalizationLayer_union1(pairs_RPN2))
checkpoint_VRD2_union2_S22_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat"+ "coordinate" information
use 6-dim feat. add Fc(6->64)w/dropout and Fc(512+64->512+6)
(LanguageModel_union(512,S=6) - LocalizationLayer_union1(pairs_RPN2))
checkpoint_VRD2_union2_S23_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat"+ "coordinate" information
use 6-dim feat. add Fc(512+6->512)
(LanguageModel_union(512,S=6) - LocalizationLayer_union1(pairs_RPN2))
checkpoint_VRD2_union2_S26_1e6_test.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat"+ "coordinate" information
use 6-dim feat. that's it
------------------------------------------------------------
(LanguageModel_att(512,S=0) - LocalizationLayer_union1(pairs_RPN))
checkpoint_VRD2_union_noAtt_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat"+ train with prev word instead of GT (w/o attention)
Must be similar to "checkpoint1e6_VRD2_union.t7"
(LanguageModel_att(512,S=0) - LocalizationLayer_union1(pairs_RPN) )
checkpoint_VRD2_union_noAtt2_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat"+ that feeds every time (w/o attention)
(LanguageModel_att(512,S=0) - LocalizationLayer_union1(pairs_RPN) )
checkpoint_VRD2_union_noAtt3_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat"+ that feeds every time (w/o attention)
input word is GT words!
--------------------------------------------------------------------
(LanguageModel_att2(S=0) - LocalizationLayer_union1(pairs_RPN) )
checkpoint_VRD2_union_Att_1e6.t7
=>learning rate 1e-6 (batch 32) on VRD dataset
use "union feat"+ that feeds every time (w/ attention)
=>require lots of memory, doesnt converge smooth
(LanguageModel_att2(S=0) - LocalizationLayer_union1(pairs_RPN) )
checkpoint_VRD2_union_Att_1e6_2.t7
=>learning rate 1e-6 (batch 16) on VRD dataset
use "union feat"+ that feeds every time (w/ attention)
checkpoint_VRD2_union_Att_1e7.t7
(LanguageModel_att2(S=whatever) - LocalizationLayer_union1(pairs_RPN3) )
checkpoint_VRD2_union_Att2_1e6.t7
=>learning rate 1e-6 (batch 32) on VRD dataset
use "union feat"+ that feeds every time (w/ attention)
starts from subj box
checkpoint_VRD2_union_Att2_1e6_2.t7
=>too fuszzy and cannot converge
(LanguageModel_att(S=whatever, ATT=true) - LocalizationLayer_union1(pairs_RPN3) )
checkpoint_VRD2_union_Att3_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat"+ that feeds every time (w/ attention)
dealing with subj+obj area
---------------------------------------------------------------------
(VRD2.h5 ,LanguageModel_att3 - LocalizationLayer_union1(pairs_RPN3) )
checkpoint_VRD2_union_clsAtt_1e6.t7
=>learning rate 1e-6 (batch 32) on VRD2 dataset
use "union feat"+ that feeds every time (w/ attention)
add cls loss and and cls based attention
checkpoint_VRD2_union_clsAtt_1e7.t7
------------------------------------------------------------------------
(VRD2.h5, cls len 17 ,LanguageModel_union3(512,S=0,length=15) - LocalizationLayer_union1(pairs_RPN) - Pairs_whatever))
checkpoint1e6_VRD2_union_MTL.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
union feat + POS classification(current word)
(VRD2.h5, cls len 17 ,LanguageModel_union3(512,S=0,length=15) - LocalizationLayer_union1(pairs_RPN) - Pairs_whatever))
checkpoint1e6_VRD2_union_MTL2.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
union feat + POS classification(future word)
(VRD2.h5, cls len 17 ,LanguageModel_union4(512,S=0,length=15,W*2) - LocalizationLayer_union1(pairs_RPN) - Pairs_whatever))
checkpoint_VRD2_union_MTL_1e6_Real.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
union feat + POS classification(current word)
POS is fed again to the model.
(VRD2.h5, cls len 17 ,LanguageModel_union3(512,S=0,length=15) - LocalizationLayer_union1(pairs_RPN))
checkpoint_VRD2_union_MTL_base_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
union feat + POS classification(current word)
input : FC from scratch
----------------------------------------------------------------
LanguageModel_union(FC(D->W)) - LocalizationLayer_union1(pairs_RPN) )
checkpoint_VRD2_union_1e6_Real.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat" only with FC7 feature as input (will be #1 baseline)
LanguageModel_union(FC(D->W)) - LocalizationLayer_union1(pairs_RPN) )
checkpoint_VRD2_union2_1e6_Real.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat" only with FC7 feature as input, add recog_base for FC7
(LanguageModel_union(512*3,S=6) - LocalizationLayer_union2(pairs_RPN2) - Pairs_whatever)
checkpoint_VRD2_union_S2thin_1e6_Real.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat+subj+obj"+ "coordinate" information
(1536*128+128*512 instead of 1536*512)
use 6-dim feat. add Fc(6->64)w/dropout and Fc(512+64->512)
--------------------------------------------------------------
(LanguageModel_dLSTM(S=6) - LocalizationLayer_union1(pairs_RPN2))
checkpoint_VRD_dLSTM_1e6_Real.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use double LSTM!
combine "union feat"+ "coordinate" information
use 6-dim feat.add Fc(6->64) add Fc(512+64->512)
------------------------------------------------------------------------------
(LanguageModel_tLSTM(S=0) - LocalizationLayer_union3(pairs_RPN), UnionSlicer)
checkpoint_VRD_tLSTM_1e6_Real.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information
.add Fc(512*3->512)
use FC7 from scratch
(LanguageModel_tLSTM(S=0) - LocalizationLayer_union3(pairs_RPN), UnionSlicer)
checkpoint_VRD_tLSTM_1e6_Real2.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information
add Fc(512*3->V directly)
use FC7 from scratch(w/o dropout)
(LanguageModel_tLSTM(S=0) - LocalizationLayer_union3(pairs_RPN), UnionSlicer)
checkpoint_VRD_tLSTM_1e6_Real3.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information
add Fc(512*3->V directly)
use spatial ave pooling for union
(LanguageModel_tLSTM(S=0) - LocalizationLayer_union3(pairs_RPN), UnionSlicer)
checkpoint_VRD_tLSTM_1e6_REAL.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information
add Fc(512*3->512)-ReLU-(512->V)
use FC((20088->512)-ReLU) for all 3
(LanguageModel_tLSTM(S=6) - LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_1e6_REAL2.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information
add Fc(512*3->512)-ReLU-(512->V)
use FC((20088->512)-ReLU) for all 3
added coordinate feat
(LanguageModel_tLSTM(S=6) - LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_1e6_REAL3.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information
add Fc(512*3->256)-ReLU-(256->V)
use FC((20088->256)-ReLU) for all 3
added coordinate feat
(LanguageModel_tLSTM(S=0) - LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_1e6_REAL4.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat
input : use 1*1conv(512->16)(ReLU) - FC((784->512)-ReLU) for all 3
output : add Fc(512*3->512)-ReLU-(512->V)
(LanguageModel_tLSTM(S=0) - LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_1e6_REAL5.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat
input : use 1*1conv(512->16)(ReLU) - FC((784->512)-ReLU) for all 3
output : (Sum pooling)-Fc(512->512)-ReLU-(512->V)
(LanguageModel_tLSTM(S=0) - LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_1e6_REAL6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (Sum pooling)-Fc(512->512)-ReLU-(512->V)
(LanguageModel_tLSTM(S=0,ATT=true) - LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_1e6_REAL7.t7
=>learning rate 1e-6 (batch 128) on VRD dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for union, recog_base for subjob
output : (attention)-Fc(512->512)-ReLU-(512->V)
(LanguageModel_tLSTM(S=0,ATT=true) - LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_1e6_REAL8.t7
=>learning rate 1e-6 (batch 128) on VRD dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for union, recog_base for subjob
output : (concat)-Fc(512*3->512)-ReLU(Dropout)-(512->V)
LanguageModel_union(FC(D->W)) - LocalizationLayer_union1(pairs_RPN) )
checkpoint_VRD2_union_1e6_REAL.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat" only with FC7(20855->512(ReLU)) feature as input
LanguageModel_union(FC(D->W)) - LocalizationLayer_union1(pairs_RPN) )
checkpoint_VRD2_union_1e6_REAL2.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat" only with 1*1conv(512->16)(ReLU) - FC((784->512)-ReLU)feature as input
LanguageModel_union(FC(D->W)) - LocalizationLayer_union1(pairs_RPN) )
checkpoint_VRD2_union_1e6_REAL3.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat" only with (avepooling) - FC((512->512)-ReLU)feature as input
(trying the first baseline again!)
----------------------------------------------------------------------------------
(LanguageModel_tLSTM(S=0,num_input=3) - LocalizationLayer_union3(pairs_RPN), UnionSlicer)
checkpoint_VRD_tLSTMbase_1e6_Real.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use triple input (no LSTM)!
combine "union feat"+ "subj"+"obj" information
.add Fc(512*3->V directly)
use spatial ave pooling for union
(LanguageModel_tLSTMbase(S=0,num_input=3) - LocalizationLayer_union3(pairs_RPN), UnionSlicer)
checkpoint_VRD_tLSTMbase_1e6_Real2.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use triple input (no LSTM)!
combine "union feat"+ "subj"+"obj" information
.add input FC((512*3->128)-ReLU-(128->512)-ReLU instead of (512*3->512))
use spatial ave pooling for union
(LanguageModel_tLSTMbase(S=0,num_input=3) - LocalizationLayer_union3(pairs_RPN), UnionSlicer)5.aijlmrtv
checkpoint_VRD_tLSTMbase_1e6_Real3.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use triple input (no LSTM)!
combine "union feat"+ "subj"+"obj" information (all ave pooling)
.add input FC((512*3->256)-ReLU-Dropout-(256->512)-ReLU instead of (512*3->512))
use spatial ave pooling for union
(LanguageModel_tLSTMbase(S=0,num_input=3) - LocalizationLayer_union3(pairs_RPN), UnionSlicer)
checkpoint_VRD_tLSTMbase_1e6_Real4.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use triple input (no LSTM)!
combine "union feat"+ "subj"+"obj" information
use avepooling for all 3
(VRD2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_MTL_1e6.t7(Deleted)
=>learning rate 1e-6 (batch 128) on VRD dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for union, recog_base for subjobj
output : (Sum pooling)-Fc(512->512)-ReLU-(512->V)
(VRD2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_MTL2_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for union, recog_base for subjob
output : (Mul pooling)-Fc(512->512)-ReLU-(512->V)
(VRD2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_MTL3_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for union, recog_base for subjob
output : (concat)-Fc(512->512)-ReLU(Dropout)-(512->V)
(VRD2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_MTL4_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for union, recog_base for subjob
output : (attention + MTL)-Fc(512->512)-ReLU-(512->V)
(VRD2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_MTL5_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for union, recog_base for subjob
output : (MTLattention,POS pooling)-Fc(512->512)-ReLU-(512->V)
(VRD2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_MTL6_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (MTLattention,POS pooling)-Fc(512->512)-ReLU-(512->V)
(VRD2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_MTL7_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (Sum)-Fc(512->512)-ReLU-(512->V)
(VRD2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VRD_tLSTM_MTL8_1e6.t7----------------------------------------(*)
=>learning rate 1e-6 (batch 128) on VRD dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (Sum)-Fc(512->512)-ReLU-(512->V)
----------------------------------------------------------------------------
( train : VRD_union.h5,eval_utils , SHUFFLE=false, LanguageModel_union(512,S=0) - LocalizationLayer_union1(pairs_RPN,SHUFFLE=false)
test : VRD.h5, SHUFFLE=true, LanguageModel_union(512,S=0) - LocalizationLayer_union1(pairs_RPN,SHUFFLE=true) )
checkpoint1e6_VRD2_union_base.t7
=>learning rate 1e-6 (batch 128) on VRD dataset (#0 baseline)
use unnion region as GT box, run DenseCap model, as it is.
=====================================================================================
( train : _union.h5,eval_utils , SHUFFLE=false, LanguageModel_union(512,S=0) - LocalizationLayer_union1(pairs_RPN,SHUFFLE=false)
test : _R.h5, SHUFFLE=true, LanguageModel_union(512,S=0) - LocalizationLayer_union1(pairs_RPN,SHUFFLE=true) )
checkpoint1e6_VG_union_base.t7
=>learning rate 1e-6 (batch 64) on VG dataset (#0 baseline)
use unnion region as GT box, run DenseCap model, as it is.
( train : _densecap.h5,eval_utils , SHUFFLE=false, LanguageModel_union(512,S=0) - LocalizationLayer_union1(pairs_RPN,SHUFFLE=false)
test : _R.h5, SHUFFLE=true, LanguageModel_union(512,S=0) - LocalizationLayer_union1(pairs_RPN,SHUFFLE=true) )
checkpoint1e6_VG_union_densecap.t7
=>learning rate 1e-6 (batch 64) on VG dataset (#0 baseline)
use unnion region as GT box, run DenseCap model, as it is.
----------------------------------------------------------------
LanguageModel_union(FC() - LocalizationLayer_union1(pairs_RPN) )
checkpoint_VG_union_0_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use "union feat" only with FC7 feature as input (will be #1 baseline)
(R2.h5, cls len 17 ,LanguageModel_union3(512,S=0,length=15) - LocalizationLayer_union1(pairs_RPN))
checkpoint_VG_union_MTL_base_1e6.t7
=>learning rate 1e-6 (batch 64) on VRD dataset
union feat + POS classification(current word)
input : FC from scratch
---------------------------------------------------------------------------------------------------
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VG_tLSTM_MTL_1e6_cntd2.t7
=>learning rate 1e-6 (batch 64) on VG dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (Sum)-Fc(512->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VG_tLSTM_MTL2_1e6_cntd.t7
=>learning rate 1e-6 (batch 64) on VG dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (concat)-Fc(512*3->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VG_tLSTM_MTL3_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : FCscratch - FC((4096->512)-ReLU) for union, recog_base for subjob
output : (concat)-Fc(512*3->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VG_tLSTM_MTL4_1e6.t7-------(in server 41)
=>learning rate 1e-6 (batch 64) on VG dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : FCscratch - FC((4096->512)-ReLU) for union, recog_base for subjob
output : (sum)-Fc(512->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VG_tLSTM_MTL5_1e6.t7-------(in server 41)
=>learning rate 1e-6 (batch 64) on VG dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat+MTL
fincune CNN
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (concat)-Fc(512*3->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VG_tLSTM_MTL6_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat+MTL
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (Sum)-Fc(512->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VG_tLSTM_MTL_weighted_1e6_cntd2.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
weighting more on long sentences (sigmoid 0~2, 5.5)
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (Sum)-Fc(512->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VG_tLSTM_MTL_weighted2_1e6_cntd.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
weighting more on long sentences (data distribution, log scale)
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (Sum)-Fc(512->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VG_tLSTM_MTL_weighted3_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
weighting more on long sentences (data distribution, log scale, max 1)
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (Sum)-Fc(512->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VG_tLSTM_MTL2_1e6_weighted2.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
weighting more on long sentences (data distribution, log scale)
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (concat)-Fc(512*3->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VG_tLSTM_MTL2_1e6_weighted3.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
weighting more on long sentences (data distribution, log scale)
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (concat)-Fc(512*3->512)-ReLU-(512->V)
-----------------------------------------------------------------------------------
(LanguageModel2 - LocalizationLayer - Pairs2)
checkpoint_VG_subjobjcoor_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use subjobd + coordinate information
(LanguageModel3 - LocalizationLayer - Pairs2)
checkpoint_VG_subjobjcoor_MTL_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use subjobd + coordinate information
-----------------------------------------------------------------------------------
(LanguageModel_union(512*3,S=0)- LocalizationLayer_union2(pairs_RPN))
checkpoint_VG_union2_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat+subj+obj" feat as input
input : ave-poooling
(LanguageModel_union(512*3,S=0)- LocalizationLayer_union2(pairs_RPN))
checkpoint_VG_union2dropout_1e6_REE.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat+subj+obj"(dropout) feat as input
input : ave-poooling
(LanguageModel_union(512*3,S=0)- LocalizationLayer_union3(pairs_RPN))
checkpoint_VG_union2scratch_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat+subj+obj"(dropout) feat as input
input : ave-poooling
(LanguageModel_union3(512*3,S=0)- LocalizationLayer_union2(pairs_RPN))
checkpoint_VG_union2_MTL_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use "union feat+subj+obj" feat as input
input : ave-poooling
-----------------------------------------------------------------------------------
(LanguageModel_union(512,S=6) - LocalizationLayer_union1(pairs_RPN2))
checkpoint_VG_union_coor_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use "union feat"+ "coordinate" information
use 6-dim feat. add Fc(6->64) and Fc(512+64->512)
(LanguageModel_union(512,S=6) - LocalizationLayer_union1(pairs_RPN2))
checkpoint_VG_union_coor2_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use "union feat"+ "coordinate" information
use 6-dim feat. add Fc(6->64) and Fc(512+64->512)
(LanguageModel_union(512,S=6) - LocalizationLayer_union1(pairs_RPN2))
checkpoint_VG_union_coor3_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use "union feat"+ "coordinate" information
use 6-dim feat. add Fc(6->64)dropout and Fc(512+64->512)
(LanguageModel_union(512,S=6) - LocalizationLayer_union1(pairs_RPN2))
checkpoint_VG_union_coor4_1e6.t7-------------------------(1)
=>learning rate 1e-6 (batch 64) on VG dataset
use "union feat"+ "coordinate" information
use 6-dim feat. add Fc(6->64)dropout and Fc(512+64->512)dropout
(LanguageModel_union(512,S=6) - LocalizationLayer_union1(pairs_RPN2))
checkpoint_VG_union_coor5_1e6.t7-------------------------(0)
=>learning rate 1e-6 (batch 64) on VG dataset
use "union feat"+ "coordinate" information
use 6-dim feat. add Fc(512+6->512)dropout
-----------------------------------------------------------------------------------
(R2.h5, cls len 17 ,LanguageModel_tLSTM2- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VG_tLSTM_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (Sum)-Fc(512->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VG_tLSTM2_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset--(will be final model)
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (concat)-Fc(512*3->512)-ReLU-(512->V)
(LanguageModel_tLSTMbase(S=0,num_input=3) - LocalizationLayer_union3(pairs_RPN), UnionSlicer)
checkpoint_VG_tLSTMbase_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use triple input (no LSTM)!
combine "union feat"+ "subj"+"obj" information
use scratch for union all 3
==================================--VG long dataset -----===================================
(make lone caption labels for VG relationship data)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VGlongv3_tLSTM_MTL_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (Sum)-Fc(512->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VGlongv3_tLSTM_MTL2_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : ave pooling - FC((512->512)-ReLU) for all 3
output : (concat)-Fc(512*3->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VGlongv3_tLSTM_MTL12_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : scratch - FC((4096->512)-ReLU) for all 3
output : (Sum)-Fc(512->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VGlongv3_tLSTM_MTL22_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : scratch - FC((4096->512)-ReLU) for all 3
output : (concat)-Fc(512*3->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VGlongv3_tLSTM_MTL13_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : scratch(w/dropout) - FC((4096->512)-ReLU) for all 3
output : (Sum)-Fc(512->512)-ReLU-(512->V)
(R2.h5, cls len 17 ,LanguageModel_tLSTM2(S=0,length=15)- LocalizationLayer_union3(pairs_RPN2), UnionSlicer)
checkpoint_VGlongv3_tLSTM_MTL23_1e6.t7
=>learning rate 1e-6 (batch 64) on VG dataset
use triple LSTM!
combine "union feat"+ "subj"+"obj" information w/ coordinate feat + MTL
input : scratch(w/ dropout) - FC((4096->512)-ReLU) for all 3
output : (concat)-Fc(512*3->512)-ReLU-(512->V)
------------------------------------------
(LanguageModel_union(512*3,S=0)- LocalizationLayer_union2(pairs_RPN))------(1)
checkpoint_VGlongv3_union2_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat+subj+obj" feat as input
input : ave-poooling
(LanguageModel_union(512*3,S=0)- LocalizationLayer_union2(pairs_RPN))---------(0)
checkpoint_VGlongv3_union2dropout_1e6.t7
=>learning rate 1e-6 (batch 128) on VRD dataset
use "union feat+subj+obj"(dropout) feat as input
input : ave-poooling