-
Notifications
You must be signed in to change notification settings - Fork 0
/
predict.py
168 lines (137 loc) · 5.86 KB
/
predict.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
# this script discretizes the whole dataset
import os
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "0,1,2,3"
import pandas as pd
import numpy as np
import argparse
import json
import torch
import time
from tqdm import tqdm
from gesture_dataset import GestureDataset
from torch.utils.data import Dataset, DataLoader
from torch.utils.tensorboard import SummaryWriter
# import viz_utils
from model import MyEncoder, MyDecoder, EncodecModel
from qt import ResidualVectorQuantizer, DummyQuantizer
# from gpt2_more_inputs import GPT2LMHeadModel
from collections import defaultdict
import itertools
def batch_to_device(batch, device):
batch_dict = {key: batch[key].to(device) for key in batch}
return batch_dict
sample_rate = 10 # cfg.sample_rate
channels = 130 # cfg.channels
seed = 42
max_sample_rate = 15
max_channels = 130
# load dfs
datamount_path = ".../kaggle-asl-fingerspelling-1st-place-solution/datamount"
df = pd.read_csv(os.path.join(datamount_path, "train_folded_real_lens.csv"))
df['is_sup'] = 0
# take fold 2 as a validation data
train_df = df[(df["fold"] != 3) & (df["fold"] != 2)].copy()
val_df = df[df["fold"] == 2].copy()
# this is the same for both train and validation
with open(os.path.join(datamount_path, 'character_to_prediction_index.json'), "r") as f:
char_to_num = json.load(f)
rev_character_map = {j: i for i, j in char_to_num.items()}
n = len(char_to_num)
pad_token = 'P'
start_token = 'S'
end_token = 'E'
char_to_num[pad_token] = n
char_to_num[start_token] = n + 1
char_to_num[end_token] = n + 2
num_to_char = {j: i for i, j in char_to_num.items()}
chars = np.array([num_to_char[i] for i in range(len(num_to_char))])
kaggle_gesture_cfg_val = {
'min_seq_len': 15,
'data_folder': '.../train_landmarks_npy_even_less/',
'symmetry_fp': os.path.join(datamount_path, 'symmetry.csv'),
'max_len': 384,
'flip_aug': 0.0,
'outer_cutmix_aug': 0.0,
'max_phrase': 31 + 2,
'pad_token': 'P',
'start_token': 'S',
'end_token': 'E',
'tokenizer': [char_to_num, num_to_char, chars]
}
parser = argparse.ArgumentParser()
args = parser.parse_args()
args.num_gpus = torch.cuda.device_count()
args.batch_size = int(36)
args.train_df = train_df
args.val_df = val_df
args.batch_size = int(args.batch_size * args.num_gpus)
args.kaggle_gesture_cfg_val = kaggle_gesture_cfg_val
device_id = 'cuda:0'
batch_size = 32
train_ds = GestureDataset(args.train_df, cfg=args.kaggle_gesture_cfg_val, mode='test')
val_ds = GestureDataset(args.val_df, cfg=args.kaggle_gesture_cfg_val, mode='test')
train_loader = DataLoader(train_ds, batch_size=batch_size, num_workers=4, collate_fn=None, shuffle=False)
val_loader = DataLoader(val_ds, batch_size=batch_size, num_workers=4, collate_fn=None, shuffle=False)
X_CHANNELS = 42
encoder = MyEncoder(channels=X_CHANNELS, dimension=128)
decoder = MyDecoder(channels=X_CHANNELS, dimension=128)
quantizer = ResidualVectorQuantizer(dimension=encoder.dimension, bins=1024, n_q=4)
model = EncodecModel(encoder, decoder, quantizer, frame_rate=10, sample_rate=10, channels=X_CHANNELS)
model.to(device_id)
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model)
model.load_state_dict(torch.load('trained_model_from_step2.pt'), strict=False) # map_location={'cuda:0': 'cuda:1'}
model.train()
with torch.no_grad():
for batch in itertools.islice(val_loader, 50):
del (batch['phrase'])
batch = batch_to_device(batch, device_id)
model(batch)
model.eval()
print(f"Number of params: {sum(p.numel() for p in model.parameters())}")
generate_codes = True
generate_embeddings = False
if generate_codes:
# save discrete codes
model.eval()
all_discrete_states = list()
all_gts = list()
all_preds = list()
all_phrases = list()
with torch.no_grad():
for ix, batch in (pbar := tqdm(enumerate(val_loader), total=len(val_loader))):
all_phrases.append(batch['phrase'])
del (batch['phrase'])
batch = batch_to_device(batch, device=device_id)
# THIS SLICES THE BATCH INTO EQUAL SLICES oF LENGTH 64
########################################################
chunked = batch['input'].unfold(1, 64, 64)
chunked_mask = batch['input_mask'].unfold(1, 64, 64)
chunked_list = [chunked[:,i].permute(0, 3, 1, 2) for i in range(6)]
chunked_mask_list = [chunked_mask[:,i] for i in range(6)]
chunked_results = []
for _inp, _mask in zip(chunked_list, chunked_mask_list):
new_batch = {'input': _inp, 'input_mask': _mask}
discrete_pred_nano = model.encode_to_discrete(new_batch).detach().cpu()
chunked_results.append(discrete_pred_nano)
# I think it worked
glued_results = torch.cat(chunked_results, dim=2)
batch['input'] = batch['input'] # [:,:60]
batch['input_mask'] = batch['input_mask'] # [:,:60]
discrete_pred = model.encode_to_discrete(batch).detach()
emb = model.quantizer.decode(discrete_pred).cpu()
discrete_pred = discrete_pred.cpu()
qres, feats_predicted = model(batch)
y_pred = qres.x
for iix in range(batch['input'].shape[0]):
gt = batch['input'][iix].cpu()
predicted = y_pred[iix].squeeze().cpu()
mask = batch['input_mask'][iix].bool().cpu()
discrete_codes = discrete_pred[iix][:, mask]
to_save_gt = gt[mask].cpu()
to_save_pred = predicted[mask]
all_discrete_states.append(discrete_codes)
all_preds.append(to_save_pred)
all_gts.append(to_save_gt)
validation_df = pd.DataFrame({'discretes': all_discrete_states, 'preds': all_preds, 'gts': all_gts, 'phrases': [j for i in all_phrases for j in i]})
validation_df.to_pickle('discrete_dataset.pkl')