-
Notifications
You must be signed in to change notification settings - Fork 17
/
Makefile
191 lines (161 loc) · 7.39 KB
/
Makefile
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
### Directory Setting
IN_DIR=input
OUT_DIR=output
META_DIR=$(OUT_DIR)/metadata
FEAT_DIR=$(OUT_DIR)/feature
MODEL_DIR=$(OUT_DIR)/model
MSRVTT2016_DIR=$(IN_DIR)/msrvtt
MSRVTT2017_DIR=$(IN_DIR)/msrvtt2017
YT2T_DIR=$(IN_DIR)/yt2t
SPLITS=train val test
DATASETS=msrvtt# yt2t msrvtt2017 tvvtt
WORD_COUNT_THRESHOLD?=3 # in output/metadata this threshold was 0; was 3 in output/metadata2017
MAX_SEQ_LEN?=30 # in output/metadata seqlen was 20; was 30 in output/metadata2017
GID?=5
DATASET?=msrvtt
TRAIN_DATASET?=$(DATASET)
VAL_DATASET?=$(DATASET)
TEST_DATASET?=$(DATASET)
TRAIN_SPLIT?=train
VAL_SPLIT?=val
TEST_SPLIT?=test
LEARNING_RATE?=0.0001
LR_UPDATE?=200
BATCH_SIZE?=64
TRAIN_SEQ_PER_IMG?=20
TEST_SEQ_PER_IMG?=20
RNN_SIZE?=512
PRINT_INTERVAL?=20
MAX_PATIENCE?=50 # FOR EARLY STOPPING
SAVE_CHECKPOINT_FROM?=1
MAX_EPOCHS?=200
NUM_CHUNKS?=1
PRINT_ATT_COEF?=0
BEAM_SIZE?=5
TODAY=20170831
EXP_NAME?=exp_$(DATASET)_$(TODAY)
VAL_LANG_EVAL?=1
TEST_LANG_EVAL?=1
EVAL_METRIC?=CIDEr
START_FROM?=No
MODEL_TYPE?=concat
POOLING?=mp
CAT_TYPE=glove
LOGLEVEL?=INFO
SS_MAX_PROB?=0.25
USE_CST?=0
SCB_CAPTIONS?=20
SCB_BASELINE?=1
USE_RL?=0
USE_RL_AFTER?=0
USE_EOS?=0
USE_MIXER?=0
MIXER_FROM?=-1
SS_K?=100
FEAT1?=resnet
FEAT2?=c3d
FEAT3?=mfcc
FEAT4?=category
FEATS?=$(FEAT1) $(FEAT2) $(FEAT3) $(FEAT4)
TRAIN_ID=$(TRAIN_DATASET)_$(MODEL_TYPE)_$(EVAL_METRIC)_$(BATCH_SIZE)_$(LEARNING_RATE)
###################################################################################################################
###
pre_process: standalize_datainfo preprocess_datainfo build_vocab create_sequencelabel convert_datainfo2cocofmt
### Standalize data
standalize_datainfo: $(foreach d,$(DATASETS),$(patsubst %,$(META_DIR)/$(d)_%_datainfo.json,$(SPLITS)))
$(META_DIR)/msrvtt_%_datainfo.json: $(MSRVTT2016_DIR)/%_videodatainfo.json
python standalize_format.py $^ $@ --dataset msrvtt2016 --split $*
$(META_DIR)/msrvtt2017_%_datainfo.json: $(MSRVTT2017_DIR)/msrvtt2017_%_videodatainfo.json
python standalize_format.py $^ $@ --dataset msrvtt2017 --split $* \
--val2016_json $(MSRVTT2016_DIR)/val_videodatainfo.json
$(META_DIR)/yt2t_%_datainfo.json: $(YT2T_DIR)/naacl15/sents_%_lc_nopunc.txt
python standalize_format.py $^ $@ --dataset yt2t
$(META_DIR)/tvvtt_%_datainfo.json: $(META_DIR)/v2t2017_infos.json
python standalize_format.py $^ $@ --dataset tvvtt --split $*
###
preprocess_datainfo: $(foreach s,$(SPLITS),$(patsubst %,$(META_DIR)/%_$(s)_proprocessedtokens.json,$(DATASETS)))
%_proprocessedtokens.json: %_datainfo.json
python preprocess_datainfo.py $^ $@
###
build_vocab: $(patsubst %,$(META_DIR)/%_train_vocab.json,$(DATASETS))
%_train_vocab.json: %_train_proprocessedtokens.json
python build_vocab.py $< $@ --word_count_threshold $(WORD_COUNT_THRESHOLD)
###
create_sequencelabel: $(foreach s,$(SPLITS),$(patsubst %,$(META_DIR)/%_$(s)_sequencelabel.h5,$(DATASETS)))
.SECONDEXPANSION:
%_sequencelabel.h5: $$(firstword $$(subst _, ,$$@))_train_vocab.json %_proprocessedtokens.json
python create_sequencelabel.py $^ $@ --max_length $(MAX_SEQ_LEN)
### Convert standalized datainfo to coco format for language evaluation
convert_datainfo2cocofmt: $(foreach s,$(SPLITS),$(patsubst %,$(META_DIR)/%_$(s)_cocofmt.json,$(DATASETS)))
%_cocofmt.json: %_datainfo.json
python convert_datainfo2cocofmt.py $< $@
### pre-compute document frequency for computing CIDEr of on model samples
compute_ciderdf: $(foreach s,$(SPLITS),$(patsubst %,$(META_DIR)/%_$(s)_ciderdf.pkl,$(DATASETS)))
%_ciderdf.pkl: %_proprocessedtokens.json
python compute_ciderdf.py $^ $@ --output_words --vocab_json $(firstword $(subst _, ,$@))_train_vocab.json
### pre-compute evaluation scores (BLEU_4, CIDEr, METEOR, ROUGE_L)
compute_evalscores: $(patsubst %,$(META_DIR)/$(TRAIN_DATASET)_%_evalscores.pkl,$(SPLITS))
%_evalscores.pkl: %_cocofmt.json
python compute_scores.py $^ $@ --remove_in_ref
#####################################################################################################################
noop=
space=$(noop) $(noop)
TRAIN_OPT=--beam_size $(BEAM_SIZE) --max_patience $(MAX_PATIENCE) --eval_metric $(EVAL_METRIC) --print_log_interval $(PRINT_INTERVAL)\
--language_eval $(VAL_LANG_EVAL) --max_epochs $(MAX_EPOCHS) --rnn_size $(RNN_SIZE) \
--train_seq_per_img $(TRAIN_SEQ_PER_IMG) --test_seq_per_img $(TEST_SEQ_PER_IMG) \
--batch_size $(BATCH_SIZE) --test_batch_size $(BATCH_SIZE) --learning_rate $(LEARNING_RATE) --lr_update $(LR_UPDATE) \
--save_checkpoint_from $(SAVE_CHECKPOINT_FROM) --num_chunks $(NUM_CHUNKS) \
--train_cached_tokens $(META_DIR)/$(TRAIN_DATASET)_train_ciderdf.pkl \
--ss_k $(SS_K) --use_rl_after $(USE_RL_AFTER) --ss_max_prob $(SS_MAX_PROB) \
--use_rl $(USE_RL) --use_mixer $(USE_MIXER) --mixer_from $(MIXER_FROM) \
--use_cst $(USE_CST) --scb_captions $(SCB_CAPTIONS) --scb_baseline $(SCB_BASELINE) \
--loglevel $(LOGLEVEL) --model_type $(MODEL_TYPE) --use_eos $(USE_EOS) \
--model_file $@ --start_from $(START_FROM) --result_file $(basename $@)_test.json \
2>&1 | tee $(basename $@).log
TEST_OPT=--beam_size $(BEAM_SIZE) \
--language_eval $(VAL_LANG_EVAL) \
--test_seq_per_img $(TEST_SEQ_PER_IMG) \
--test_batch_size $(BATCH_SIZE) \
--loglevel $(LOGLEVEL) \
--result_file $@
train: $(MODEL_DIR)/$(EXP_NAME)/$(subst $(space),$(noop),$(FEATS))_$(TRAIN_ID).pth
$(MODEL_DIR)/$(EXP_NAME)/$(subst $(space),$(noop),$(FEATS))_$(TRAIN_ID).pth: \
$(META_DIR)/$(TRAIN_DATASET)_$(TRAIN_SPLIT)_sequencelabel.h5 \
$(META_DIR)/$(VAL_DATASET)_$(VAL_SPLIT)_sequencelabel.h5 \
$(META_DIR)/$(TEST_DATASET)_$(TEST_SPLIT)_sequencelabel.h5 \
$(META_DIR)/$(TRAIN_DATASET)_$(TRAIN_SPLIT)_cocofmt.json \
$(META_DIR)/$(VAL_DATASET)_$(VAL_SPLIT)_cocofmt.json \
$(META_DIR)/$(TEST_DATASET)_$(TEST_SPLIT)_cocofmt.json \
$(META_DIR)/$(TRAIN_DATASET)_$(TRAIN_SPLIT)_evalscores.pkl \
$(patsubst %,$(FEAT_DIR)/$(TRAIN_DATASET)_$(TRAIN_SPLIT)_%_mp$(NUM_CHUNKS).h5,$(FEATS)) \
$(patsubst %,$(FEAT_DIR)/$(VAL_DATASET)_$(VAL_SPLIT)_%_mp$(NUM_CHUNKS).h5,$(FEATS)) \
$(patsubst %,$(FEAT_DIR)/$(TEST_DATASET)_$(TEST_SPLIT)_%_mp$(NUM_CHUNKS).h5,$(FEATS))
mkdir -p $(MODEL_DIR)/$(EXP_NAME)
CUDA_VISIBLE_DEVICES=$(GID) python train.py \
--train_label_h5 $(word 1,$^) \
--val_label_h5 $(word 2,$^) \
--test_label_h5 $(word 3,$^) \
--train_cocofmt_file $(word 4,$^) \
--val_cocofmt_file $(word 5,$^) \
--test_cocofmt_file $(word 6,$^) \
--train_bcmrscores_pkl $(word 7,$^) \
--train_feat_h5 $(patsubst %,$(FEAT_DIR)/$(TRAIN_DATASET)_$(TRAIN_SPLIT)_%_mp$(NUM_CHUNKS).h5,$(FEATS))\
--val_feat_h5 $(patsubst %,$(FEAT_DIR)/$(VAL_DATASET)_$(VAL_SPLIT)_%_mp$(NUM_CHUNKS).h5,$(FEATS))\
--test_feat_h5 $(patsubst %,$(FEAT_DIR)/$(TEST_DATASET)_$(TEST_SPLIT)_%_mp$(NUM_CHUNKS).h5,$(FEATS))\
$(TRAIN_OPT)
test: $(MODEL_DIR)/$(EXP_NAME)/$(subst $(space),$(noop),$(FEATS))_$(TRAIN_ID)_test.json
$(MODEL_DIR)/$(EXP_NAME)/$(subst $(space),$(noop),$(FEATS))_$(TRAIN_ID)_test.json: \
$(MODEL_DIR)/$(EXP_NAME)/$(subst $(space),$(noop),$(FEATS))_$(TRAIN_ID).pth \
$(META_DIR)/$(TEST_DATASET)_$(TEST_SPLIT)_sequencelabel.h5 \
$(META_DIR)/$(TEST_DATASET)_$(TEST_SPLIT)_cocofmt.json \
$(patsubst %,$(FEAT_DIR)/$(TEST_DATASET)_$(TEST_SPLIT)_%_mp$(NUM_CHUNKS).h5,$(FEATS))
CUDA_VISIBLE_DEVICES=$(GID) python test.py \
--model_file $(word 1,$^) \
--test_label_h5 $(word 2,$^) \
--test_cocofmt_file $(word 3,$^) \
--test_feat_h5 $(patsubst %,$(FEAT_DIR)/$(TEST_DATASET)_$(TEST_SPLIT)_%_mp$(NUM_CHUNKS).h5,$(FEATS))\
$(TEST_OPT)
# You can use the wildcard with .PRECIOUS.
.PRECIOUS: %.pth
# If you want all intermediates to remain
.SECONDARY: