forked from facebookresearch/vilbert-multi-task
-
Notifications
You must be signed in to change notification settings - Fork 0
/
vilbert_tasks.yml
351 lines (351 loc) · 9.42 KB
/
vilbert_tasks.yml
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
TASK1:
name: VQA
type: VL-classifier
loss: BCEWithLogitLoss
process: normal
task_id: 1
dataroot: datasets/VQA/
features_h5path1: datasets/coco/features_100/COCO_trainval_resnext152_faster_rcnn_genome.lmdb
features_h5path2: ''
train_annotations_jsonpath: ''
val_annotations_jsonpath: ''
max_seq_length: 23
max_region_num: 101
batch_size: 128
eval_batch_size: 1024
train_split: trainval
val_split: minval
lr: 0.00004
num_epoch: 20
TASK2:
name: GenomeQA
type: VL-classifier
loss: BCEWithLogitLoss
process: normal
task_id: 2
dataroot: datasets/visual_genome/
features_h5path1: datasets/visual_genome/vg_resnext152_faster_rcnn_genome.lmdb
features_h5path2: ''
train_annotations_jsonpath: ''
val_annotations_jsonpath: ''
max_seq_length: 26
max_region_num: 101
batch_size: 128
eval_batch_size: 1024
train_split: train
val_split: val
lr: 0.00004
num_epoch: 20
TASK3:
name: VisualDialog
type: VL-logit
loss: CrossEntropyLoss
process: dialog
task_id: 3
dataroot: datasets/visual_dialog
features_h5path1: datasets/coco/features_100/COCO_trainval_resnext152_faster_rcnn_genome.lmdb
features_h5path2: ''
train_annotations_jsonpath: datasets/visual_dialog/visdial_1.0_train.json
val_annotations_jsonpath: datasets/visual_dialog/visdial_1.0_val.json
max_seq_length: 16
max_region_num: 101
batch_size: 64
train_split: train
val_split: val
lr: 0.00004
num_epoch: 20
TASK4:
name: Visual7w
type: V-logit-mc
loss: BCEWithLogitLoss
process: normal
task_id: 4
dataroot: datasets/visual7w
features_h5path1: datasets/visual7w/visual7w_resnext152_faster_rcnn_genome.lmdb
features_h5path2: datasets/visual7w/visual7w_gt_resnext152_faster_rcnn_genome.lmdb
train_annotations_jsonpath: ''
val_annotations_jsonpath: ''
max_seq_length: 20
max_region_num: 200
batch_size: 256
train_split: train
val_split: val
lr: 0.00002
num_epoch: 20
TASK5:
name: VCR_Q-A #visual commonsense reasoning QA (only answering I think)
type: VL-logit
loss: CrossEntropyLoss
process: expand
task_id: 5
dataroot: datasets/VCR
features_h5path1: datasets/VCR/VCR_resnext152_faster_rcnn_genome.lmdb
features_h5path2: datasets/VCR/VCR_gt_resnext152_faster_rcnn_genome.lmdb
train_annotations_jsonpath: datasets/VCR/train.jsonl
val_annotations_jsonpath: datasets/VCR/val.jsonl
max_seq_length: 60
max_region_num: 101
batch_size: 64
train_split: train
val_split: val
lr: 0.00002
num_epoch: 10
TASK6:
name: VCR_QA-R
type: VL-logit
loss: CrossEntropyLoss
process: expand
task_id: 6
dataroot: datasets/VCR
features_h5path1: datasets/VCR/VCR_resnext152_faster_rcnn_genome.lmdb
features_h5path2: datasets/VCR/VCR_gt_resnext152_faster_rcnn_genome.lmdb
train_annotations_jsonpath: datasets/VCR/train.jsonl
val_annotations_jsonpath: datasets/VCR/val.jsonl
max_seq_length: 80
max_region_num: 101
batch_size: 64
train_split: train
val_split: val
lr: 0.00002
num_epoch: 10
TASK7:
name: RetrievalCOCO
type: VL-logit
loss: CrossEntropyLoss
process: retrieval
task_id: 7
dataroot: datasets/cocoRetreival
features_h5path1: datasets/coco/features_100/COCO_trainval_resnext152_faster_rcnn_genome.lmdb
features_h5path2: ''
train_annotations_jsonpath: datasets/cocoRetreival/all_data_final_train_2014.jsonline
val_annotations_jsonpath: datasets/cocoRetreival/all_data_final_test_set0_2014.jsonline
max_seq_length: 30
max_region_num: 101
batch_size: 128
train_split: train
val_split: val
lr: 0.00002
num_epoch: 20
TASK8:
name: RetrievalFlickr30k
type: VL-logit
loss: CrossEntropyLoss
process: retrieval
task_id: 8
dataroot: datasets/flickr30k
features_h5path1: datasets/flickr30k/flickr30k_resnext152_faster_rcnn_genome.lmdb
features_h5path2: ''
train_annotations_jsonpath: datasets/flickr30k/all_data_final_train_2014.jsonline
val_annotations_jsonpath: datasets/flickr30k/all_data_final_test_set0_2014.jsonline
max_seq_length: 30
max_region_num: 101
batch_size: 128
train_split: train
val_split: val
lr: 0.00002
num_epoch: 20
TASK9:
name: refcoco
type: V-logit
loss: BCEWithLogitLoss
process: normal
task_id: 9
dataroot: datasets/refcoco
features_h5path1: datasets/refcoco/refcoco_unc/refcoco_resnext152_faster_rcnn_genome.lmdb
features_h5path2: datasets/refcoco/refcoco_unc/refcoco_gt_resnext152_faster_rcnn_genome.lmdb
train_annotations_jsonpath: ''
val_annotations_jsonpath: ''
max_seq_length: 20
max_region_num: 101
batch_size: 256
train_split: train
val_split: val
lr: 0.00002
num_epoch: 20
TASK10:
name: refcoco+
type: V-logit
loss: BCEWithLogitLoss
process: normal
task_id: 10
dataroot: datasets/refcoco
features_h5path1: datasets/refcoco/refcoco+_unc/refcoco+_resnext152_faster_rcnn_genome.lmdb
features_h5path2: datasets/refcoco/refcoco+_unc/refcoco+_gt_resnext152_faster_rcnn_genome.lmdb
train_annotations_jsonpath: ''
val_annotations_jsonpath: ''
max_seq_length: 20
max_region_num: 101
batch_size: 256
eval_batch_size: 1024
train_split: train
val_split: val
lr: 0.00002
num_epoch: 20
TASK11:
name: refcocog # refer_expression_dataset ???
type: V-logit # why V-logit ?
loss: BCEWithLogitLoss
process: normal
task_id: 11
dataroot: datasets/refcoco
features_h5path1: datasets/refcoco/refcocog_umd/refcocog_resnext152_faster_rcnn_genome.lmdb
features_h5path2: datasets/refcoco/refcocog_umd/refcocog_gt_resnext152_faster_rcnn_genome.lmdb
train_annotations_jsonpath: ''
val_annotations_jsonpath: ''
max_seq_length: 20
max_region_num: 101
batch_size: 256
train_split: train
val_split: val
start_iteration: 0
lr: 0.00002
num_epoch: 20
TASK12:
name: NLVR2
type: VL-binary-classifier
loss: BCEWithLogitLoss
process: nlvr
task_id: 12
dataroot: datasets/nlvr2/
features_h5path1: datasets/nlvr2/nlvr2_resnext152_faster_rcnn_genome.lmdb
features_h5path2: ''
train_annotations_jsonpath: ''
val_annotations_jsonpath: ''
max_seq_length: 40
max_region_num: 101
batch_size: 64
eval_batch_size: 512
train_split: train
val_split: dev
lr: 0.00002
num_epoch: 20
TASK13:
name: VisualEntailment
type: VL-tri-classifier
loss: BCEWithLogitLoss
process: normal
task_id: 13
dataroot: datasets/visual_entailment/
features_h5path1: datasets/flickr30k/flickr30k_resnext152_faster_rcnn_genome.lmdb
features_h5path2: ''
train_annotations_jsonpath: ''
val_annotations_jsonpath: ''
max_seq_length: 56
max_region_num: 101
batch_size: 256
eval_batch_size: 1024
train_split: train
val_split: dev
lr: 0.00002
num_epoch: 20
TASK14:
name: GuessWhat
type: VL-tri-classifier
loss: BCEWithLogitLoss
process: normal
task_id: 14
dataroot: datasets/guesswhat/
features_h5path1: datasets/coco/features_100/COCO_trainval_resnext152_faster_rcnn_genome.lmdb
features_h5path2: ''
train_annotations_jsonpath: ''
val_annotations_jsonpath: ''
max_seq_length: 25
max_region_num: 101
batch_size: 256
eval_batch_size: 1024
train_split: train
val_split: valid
lr: 0.00004
num_epoch: 20
TASK15:
name: GQA
type: VL-classifier-GQA
loss: BCEWithLogitLoss
process: normal
task_id: 15
dataroot: datasets/gqa/
features_h5path1: datasets/gqa/gqa_resnext152_faster_rcnn_genome.lmdb
features_h5path2: ''
train_annotations_jsonpath: ''
val_annotations_jsonpath: ''
max_seq_length: 26
max_region_num: 101
batch_size: 128
eval_batch_size: 1024
train_split: trainval
val_split: minval
lr: 0.00004
num_epoch: 20
TASK16:
name: Foil
type: VL-binary-classifier
loss: CrossEntropyLoss
process: normal
task_id: 16
dataroot: datasets/Foil
features_h5path1: datasets/coco/features_100/COCO_trainval_resnext152_faster_rcnn_genome.lmdb
features_h5path2: ''
train_annotations_jsonpath: datasets/Foil/foilv1.0_train_2017.json
val_annotations_jsonpath: datasets/Foil/foilv1.0_test_2017.json
max_seq_length: 20
max_region_num: 101
batch_size: 256
eval_batch_size: 1024
train_split: train
val_split: val
lr: 0.00004
num_epoch: 20
TASK17:
name: GuessWhatPointing
type: V-logit-mc
loss: BCEWithLogitLoss
process: normal
task_id: 17
dataroot: datasets/guesswhat/
features_h5path1: datasets/coco/features_100/COCO_trainval_resnext152_faster_rcnn_genome.lmdb
features_h5path2: datasets/guesswhat/guesswhat_gt_resnext152_faster_rcnn_genome.lmdb
train_annotations_jsonpath: ''
val_annotations_jsonpath: ''
max_seq_length: 256
max_region_num: 306
batch_size: 64
train_split: train
val_split: valid
lr: 0.00002
num_epoch: 20
TASK18:
name: FlickrGrounding
type: V-logit
loss: BCEWithLogitLoss
process: normal
task_id: 18
dataroot: datasets/flickr30k/
features_h5path1: datasets/flickr30k/flickr30k_resnext152_faster_rcnn_genome.lmdb
features_h5path2: datasets/flickr30k/flickr30k_gt_resnext152_faster_rcnn_genome.lmdb
train_annotations_jsonpath: ''
val_annotations_jsonpath: ''
max_seq_length: 24
max_region_num: 200
batch_size: 256
train_split: train
val_split: val
lr: 0.000002
num_epoch: 20
TASK19:
name: ContextualCaptioning
type: VL-logit
loss: CrossEntropyLoss
process: normal
task_id: 19
dataroot: /mnt/nas2/seungil
features_h5path1: /mnt/nas2/seungil/lmdbs/FA.lmdb
features_h5path2: ''
train_annotations_jsonpath: '/mnt/nas2/seungil/jsonlines/FA.jsonline'
val_annotations_jsonpath: '/mnt/nas2/seungil/jsonlines/FA.jsonline'
max_seq_length: 256
max_region_num: 101
batch_size: 8
train_split: train
val_split: val
lr: 0.000002
num_epoch: 20