Skip to content

Commit 01a5606

Browse files
apsdehalapsdehal
authored
[refactor] Add config_path and proper config structure for dataset/model (#48)
1 parent e670c87 commit 01a5606

File tree

130 files changed

+620
-832
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

130 files changed

+620
-832
lines changed

‎configs/captioning/coco/butd.yaml

-37
Original file line numberDiff line numberDiff line change
@@ -1,40 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/captioning/coco.yaml
3-
model_config:
4-
butd: &butd
5-
model_data_dir: ../data/
6-
metrics:
7-
- type: caption_bleu4
8-
losses:
9-
- type: caption_cross_entropy
10-
classifier:
11-
type: language_decoder
12-
params:
13-
dropout: 0.5
14-
hidden_dim: 1024
15-
feature_dim: 2048
16-
fc_bias_init: 0
17-
image_feature_embeddings:
18-
- modal_combine:
19-
type: top_down_attention_lstm
20-
params:
21-
dropout: 0.5
22-
hidden_dim: 1024
23-
attention_dim: 1024
24-
normalization: softmax
25-
transform:
26-
type: linear
27-
params:
28-
out_dim: 1
29-
image_feature_dim: 2048
30-
embedding_dim: 300
31-
image_feature_encodings:
32-
- type: finetune_faster_rcnn_fpn_fc7
33-
params:
34-
bias_file: detectron/fc6/fc7_b.pkl
35-
weights_file: detectron/fc6/fc7_w.pkl
36-
inference:
37-
type: greedy
381
optimizer:
392
type: Adamax
403
params:

‎configs/captioning/m4c_textcaps/butd.yaml

+1-37
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/captioning/coco.yaml
31
dataset_config:
42
coco:
53
image_features:
@@ -43,41 +41,7 @@ dataset_config:
4341
use_ocr: false
4442
# Return spatial information of OCR tokens if present
4543
use_ocr_info: false
46-
model_config:
47-
butd: &butd
48-
model_data_dir: ../data/
49-
metrics:
50-
- type: caption_bleu4
51-
losses:
52-
- type: caption_cross_entropy
53-
classifier:
54-
type: language_decoder
55-
params:
56-
dropout: 0.5
57-
hidden_dim: 1024
58-
feature_dim: 2048
59-
fc_bias_init: 0
60-
image_feature_embeddings:
61-
- modal_combine:
62-
type: top_down_attention_lstm
63-
params:
64-
dropout: 0.5
65-
hidden_dim: 1024
66-
attention_dim: 1024
67-
normalization: softmax
68-
transform:
69-
type: linear
70-
params:
71-
out_dim: 1
72-
image_feature_dim: 2048
73-
embedding_dim: 300
74-
image_feature_encodings:
75-
- type: finetune_faster_rcnn_fpn_fc7
76-
params:
77-
bias_file: detectron/fc6/fc7_b.pkl
78-
weights_file: detectron/fc6/fc7_w.pkl
79-
inference:
80-
type: greedy
44+
8145
optimizer:
8246
type: Adamax
8347
params:

‎configs/captioning/m4c_textcaps/m4c_captioner.yaml

+1-32
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/captioning/m4c_textcaps.yaml
31
# Use soft copy
42
dataset_config:
53
m4c_textcaps:
@@ -43,36 +41,7 @@ dataset_config:
4341
type: phoc
4442
params:
4543
max_length: 50
46-
model_config:
47-
m4c_captioner:
48-
lr_scale_frcn: 0.1
49-
lr_scale_text_bert: 0.1
50-
lr_scale_mmt: 1.0 # no scaling
51-
text_bert_init_from_bert_base: true
52-
text_bert:
53-
num_hidden_layers: 3
54-
obj:
55-
mmt_in_dim: 2048
56-
dropout_prob: 0.1
57-
ocr:
58-
mmt_in_dim: 3002 # 300 (FastText) + 604 (PHOC) + 2048 (Faster R-CNN) + 50 (all zeros; legacy)
59-
dropout_prob: 0.1
60-
mmt:
61-
hidden_size: 768
62-
num_hidden_layers: 4
63-
classifier:
64-
type: linear
65-
ocr_max_num: 50
66-
ocr_ptr_net:
67-
hidden_size: 768
68-
query_key_size: 768
69-
params: {}
70-
model_data_dir: ../data
71-
metrics:
72-
- type: textcaps_bleu4
73-
losses:
74-
- type: m4c_decoding_bce_with_mask
75-
remove_unk_in_pred: true
44+
7645
optimizer:
7746
params:
7847
eps: 1.0e-08

‎configs/captioning/m4c_textcaps/m4c_captioner_coco.yaml

-2
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/captioning/m4c_textcaps.yaml
31
# Use soft copy
42
dataset_config:
53
m4c_textcaps:

‎configs/captioning/m4c_textcaps/m4c_captioner_coco_eval_on_textcaps.yaml

-2
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,3 @@
1-
includes:
2-
- ../configs/captioning/m4c_textcaps/m4c_captioner_coco.yaml
31
dataset_config:
42
m4c_textcaps:
53
image_features:

‎configs/captioning/m4c_textcaps/m4c_captioner_coco_textcaps_joint.yaml

-2
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/captioning/m4c_textcaps.yaml
31
# Use soft copy
42
dataset_config:
53
m4c_textcaps:

‎configs/captioning/m4c_textcaps/m4c_captioner_without_ocr.yaml

-2
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/captioning/m4c_textcaps.yaml
31
# Use soft copy
42
dataset_config:
53
m4c_textcaps:

‎configs/pythia_bert/cc_pretrain.yaml

-55
Original file line numberDiff line numberDiff line change
@@ -1,60 +1,5 @@
11
includes:
22
- ../configs/pythia_bert/coco_pretrain.yaml
3-
- common/defaults/configs/datasets/captioning/masked_cc.yaml
4-
5-
model_config:
6-
pythia_bert:
7-
combine_embeddings: true
8-
model_data_dir: ../data/
9-
bert_model_name: bert-base-uncased
10-
training_head_type: pretraining
11-
visual_embedding_dim: 2048
12-
special_visual_initialize: true
13-
hard_cap_seq_len: null
14-
cut_first: text
15-
embedding_strategy: plain
16-
bypass_transformer: false
17-
output_attentions: false
18-
output_hidden_states: false
19-
text_only: false
20-
random_initialize: false
21-
pretrained_bert: true
22-
image_intra_attention: true
23-
freeze_base: false
24-
finetune_lr_multiplier: 1
25-
image_feature_embeddings:
26-
- embed_dim: 768
27-
kdim: 768
28-
vdim: 768
29-
num_heads: 8
30-
dropout: 0.1
31-
image_feature_dim: 2048
32-
image_feature_projection:
33-
module: linear
34-
in_dim: 2048
35-
out_dim: 768
36-
image_feature_encodings:
37-
- type: finetune_faster_rcnn_fpn_fc7
38-
params:
39-
bias_file: /private/home/vedanuj/pythia/data/detectron/resnext152_fc6/fc7_b.pkl
40-
weights_file: /private/home/vedanuj/pythia/data/detectron/resnext152_fc6/fc7_w.pkl
41-
image_text_modal_combine:
42-
type: non_linear_element_multiply
43-
params:
44-
dropout: 0
45-
hidden_dim: 768
46-
image_feature_attentions:
47-
- embed_dim: 768
48-
kdim: 768
49-
vdim: 768
50-
num_heads: 8
51-
dropout: 0.1
52-
text_embeddings:
53-
- embed_dim: 768
54-
kdim: 768
55-
vdim: 768
56-
num_heads: 8
57-
dropout: 0.1
583

594
dataset_config:
605
masked_cc:

‎configs/pythia_bert/coco_pretrain.yaml

-57
Original file line numberDiff line numberDiff line change
@@ -1,60 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/captioning/masked_coco.yaml
3-
4-
model_config:
5-
pythia_bert:
6-
combine_embeddings: true
7-
model_data_dir: ../data/
8-
bert_model_name: bert-base-uncased
9-
training_head_type: pretraining
10-
visual_embedding_dim: 2048
11-
special_visual_initialize: true
12-
hard_cap_seq_len: null
13-
cut_first: text
14-
embedding_strategy: plain
15-
bypass_transformer: false
16-
output_attentions: false
17-
output_hidden_states: false
18-
text_only: false
19-
random_initialize: false
20-
pretrained_bert: true
21-
image_intra_attention: true
22-
freeze_base: false
23-
finetune_lr_multiplier: 1
24-
image_feature_embeddings:
25-
- embed_dim: 768
26-
kdim: 768
27-
vdim: 768
28-
num_heads: 8
29-
dropout: 0.1
30-
image_feature_dim: 2048
31-
image_feature_projection:
32-
module: linear
33-
in_dim: 2048
34-
out_dim: 768
35-
image_feature_encodings:
36-
- type: finetune_faster_rcnn_fpn_fc7
37-
params:
38-
bias_file: /private/home/vedanuj/pythia/data/detectron/resnext152_fc6/fc7_b.pkl
39-
weights_file: /private/home/vedanuj/pythia/data/detectron/resnext152_fc6/fc7_w.pkl
40-
image_text_modal_combine:
41-
type: non_linear_element_multiply
42-
params:
43-
dropout: 0
44-
hidden_dim: 768
45-
image_feature_attentions:
46-
- embed_dim: 768
47-
kdim: 768
48-
vdim: 768
49-
num_heads: 8
50-
dropout: 0.1
51-
text_embeddings:
52-
- embed_dim: 768
53-
kdim: 768
54-
vdim: 768
55-
num_heads: 8
56-
dropout: 0.1
57-
581
dataset_config:
592
masked_coco:
603
return_info: true

‎configs/pythia_bert/masked_vqa2_pretrain.yaml

-1
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,4 @@
11
includes:
2-
- common/defaults/configs/datasets/vqa/masked_vqa2.yaml
32
- ../configs/pythia_bert/coco_pretrain.yaml
43

54
dataset_config:

‎configs/pythia_bert/visual_entailment.yaml

-1
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,4 @@
11
includes:
2-
- common/defaults/configs/datasets/reasoning/visual_entailment.yaml
32
- ../configs/pythia_bert/vqa2.yaml
43

54
model_config:

‎configs/pythia_bert/vizwiz.yaml

-1
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,4 @@
11
includes:
2-
- common/defaults/configs/datasets/vqa/vizwiz.yaml
32
- ../configs/pythia_bert/vqa2.yaml
43

54
model_config:

‎configs/pythia_bert/vqa2.yaml

-1
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,4 @@
11
includes:
2-
- common/defaults/configs/datasets/vqa/vqa2.yaml
32
- ../configs/pythia_bert/coco_pretrain.yaml
43

54
dataset_config:

‎configs/vilbert/cc_pretrain.yaml

-4
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/captioning/masked_cc.yaml
3-
- common/defaults/configs/models/vilbert.yaml
4-
51
model_config:
62
vilbert:
73
training_head_type: pretraining

‎configs/vilbert/coco_pretrain.yaml

-4
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/captioning/masked_coco.yaml
3-
- common/defaults/configs/models/vilbert.yaml
4-
51
model_config:
62
vilbert:
73
training_head_type: pretraining

‎configs/vilbert/coco_pretrain_train_val.yaml

-4
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/captioning/masked_coco.yaml
3-
- common/defaults/configs/models/vilbert.yaml
4-
51
dataset_config:
62
masked_coco:
73
return_info: true

‎configs/vilbert/masked_vqa2_pretrain.yaml

-4
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/vqa/masked_vqa2.yaml
3-
- common/defaults/configs/models/vilbert.yaml
4-
51
model_config:
62
vilbert:
73
training_head_type: pretraining

‎configs/vilbert/masked_vqa2_pretrain_train_val.yaml

-4
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/vqa/masked_vqa2.yaml
3-
- common/defaults/configs/models/vilbert.yaml
4-
51
model_config:
62
vilbert:
73
training_head_type: pretraining

‎configs/vilbert/mmimdb.yaml

-4
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/reasoning/mmimdb.yaml
3-
- common/defaults/configs/models/vilbert.yaml
4-
51
model_config:
62
vilbert:
73
training_head_type: mmimdb

‎configs/vilbert/nlvr2.yaml

-4
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/reasoning/nlvr2.yaml
3-
- common/defaults/configs/models/vilbert.yaml
4-
51
model_config:
62
vilbert:
73
training_head_type: nlvr2

‎configs/vilbert/visual_entailment.yaml

-4
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/reasoning/visual_entailment.yaml
3-
- common/defaults/configs/models/vilbert.yaml
4-
51
model_config:
62
vilbert:
73
training_head_type: visual_entailment

‎configs/vilbert/vizwiz.yaml

-4
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/vqa/vizwiz.yaml
3-
- common/defaults/configs/models/vilbert.yaml
4-
51
model_config:
62
vilbert:
73
training_head_type: vizwiz

‎configs/vilbert/vqa2.yaml

-4
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/vqa/vqa2.yaml
3-
- common/defaults/configs/models/vilbert.yaml
4-
51
model_config:
62
vilbert:
73
training_head_type: vqa

‎configs/vilbert/vqa2_train_val.yaml

-4
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/vqa/vqa2.yaml
3-
- common/defaults/configs/models/vilbert.yaml
4-
51
model_config:
62
vilbert:
73
training_head_type: vqa

‎configs/visual_bert/cc_pretrain.yaml

-3
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,3 @@
1-
includes:
2-
- common/defaults/configs/datasets/captioning/masked_cc.yaml
3-
41
model_config:
52
visual_bert:
63
bert_model_name: bert-base-uncased

0 commit comments

Comments
 (0)