Vid2Sim/train_predictor.py at main · CzzzzH/Vid2Sim · GitHub

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
import sys
sys.path.append('./')

import torch
import numpy as np
import os
import wandb
import argparse

from torch import nn
from torch.utils.data import DataLoader
from transformers import VideoMAEModel
from tqdm import tqdm
from utils.seeding import seed_everything
from models.phys_predictor import SimulationDataset, FeedForwardPredictor

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

def train_predictor(args):

    if args.use_wandb:
        os.environ["WANDB_API_KEY"] = "YOUR_WANDB_API_KEY"
        wandb.login()
        wandb.init(project="simulator-feedforward")

    train_dataset = SimulationDataset(args.dataset_dir, res=args.res, frame_num=16, split='train')
    val_dataset = SimulationDataset(args.dataset_dir, res=args.res, frame_num=16, split='val')
    train_dataloader = DataLoader(train_dataset, batch_size=args.bs, num_workers=args.num_workers, shuffle=True)
    val_dataloader = DataLoader(val_dataset, batch_size=1, num_workers=args.num_workers, shuffle=False)
    backbone_model_pretrained = VideoMAEModel.from_pretrained("MCG-NJU/videomae-base")

    # Interpolate position embeddings
    if args.res != 224:
        backbone_model_pretrained.embeddings.patch_embeddings.image_size = (args.res, args.res)
        pos_tokens = backbone_model_pretrained.embeddings.position_embeddings
        T = 8
        P = int((pos_tokens.shape[1] // T) ** 0.5)
        C = pos_tokens.shape[2]
        new_P = args.res // 16
        # B, L, C -> BT, H, W, C -> BT, C, H, W
        pos_tokens = pos_tokens.reshape(-1, T, P, P, C)
        pos_tokens = pos_tokens.reshape(-1, P, P, C).permute(0, 3, 1, 2)
        pos_tokens = torch.nn.functional.interpolate(
            pos_tokens, size=(new_P, new_P), mode='bicubic', align_corners=False)
        # BT, C, H, W -> BT, H, W, C ->  B, T, H, W, C
        pos_tokens = pos_tokens.permute(0, 2, 3, 1).reshape(-1, T, new_P, new_P, C)
        pos_tokens = pos_tokens.flatten(1, 3)  # B, L, C
        backbone_model_pretrained.embeddings.position_embeddings = pos_tokens  # update

    predictor = FeedForwardPredictor(backbone_model_pretrained, in_features=768).to(device)
    if args.resume_path is not None and os.path.exists(args.resume_path):
        predictor.load_state_dict(torch.load(args.resume_path))

    loss_fn = nn.MSELoss()
    optimizer = torch.optim.Adam([{'params': predictor.backbone.parameters(), 'lr': args.lr},
                                   {'params': predictor.regression_head_yms.parameters(), 'lr': args.lr},
                                   {'params': predictor.regression_head_prs.parameters(), 'lr': args.lr},
                                   {'params': predictor.lbs_head.parameters(), 'lr': args.lr}
                                   ], betas=(0.9, 0.98))

    os.makedirs(f'{args.save_dir}/{args.task_tag}', exist_ok=True)
    predictor.train()
    total_iteration = 1

    for epoch in range(1, args.total_epochs + 1):

        pbar = tqdm(train_dataloader)

        for data in pbar:

            video = data['video'].to(device)
            yms_gt = data['yms_gt'].to(device)
            prs_gt = data['prs_gt'].to(device)
            lbs_gt = data['lbs_gt'].to(device)

            output = predictor(video)
            yms_pred, prs_pred, lbs_pred = output['yms'], output['prs'], output['lbs']

            parameter_loss = loss_fn(yms_pred, yms_gt) + 100 * loss_fn(prs_pred, prs_gt)
            lbs_loss = loss_fn(lbs_pred, lbs_gt)
            loss = parameter_loss + lbs_loss

            pbar.set_description(f'Epoch {epoch}, Total Iteration {total_iteration}, Loss: {parameter_loss.item()}')
            if args.use_wandb:
                wandb.log({'train_loss': loss.item()})

            optimizer.zero_grad()
            loss.backward(retain_graph=True)
            optimizer.step()

            # Validation
            if total_iteration % args.val_interval == 0:

                torch.save(predictor , f'{args.save_dir}/{args.task_tag}/ckpt_{total_iteration:06d}.pth')

                print("Starting Validation")
                predictor.eval()
                valid_losses = []

                for i, data in enumerate(val_dataloader):

                    video = data['video'].to(device)
                    yms_gt = data['yms_gt'].to(device)
                    prs_gt = data['prs_gt'].to(device)
                    lbs_gt = data['lbs_gt'].to(device)

                    output = predictor(video)
                    yms_pred, prs_pred, lbs_pred = output['yms'], output['prs'], output['lbs']
                    lbs_loss = loss_fn(lbs_pred, lbs_gt)
                    loss = parameter_loss + lbs_loss
                    valid_losses.append(loss.item())

                    print(f'Total iteration {total_iteration}, Loss: {loss.item()}, LBS Loss: {lbs_loss.item()} \
                        yms_gt: {torch.pow(10, yms_gt).item()}, yms_pred: {torch.pow(10, yms_pred).item()}, \
                        prs_gt: {prs_gt.item()}, prs_pred: {prs_pred.item()}')

                valid_losses_mean = np.mean(valid_losses)
                if args.use_wandb:
                    wandb.log({f'val_loss_{args.task_tag}': valid_losses_mean})
                    wandb.log({f'val_loss': valid_losses_mean})

                predictor.train()

            # Save
            if total_iteration % args.save_interval == 0:
                torch.save(predictor , f'{args.save_dir}/{args.task_tag}/ckpt_{total_iteration:06d}.pth')

            total_iteration += 1

    wandb.finish()

if __name__ == "__main__":

    parser = argparse.ArgumentParser()
    parser.add_argument("--dataset_dir", type=str, default="dataset/objaverse")
    parser.add_argument("--save_dir", type=str, default="checkpoints")
    parser.add_argument("--use_wandb", type=bool, default=False)
    parser.add_argument("--task_tag", type=str, default="phys_predictor")
    parser.add_argument("--res", type=int, default=448)
    parser.add_argument("--bs", type=int, default=4)
    parser.add_argument("--num_workers", type=int, default=8)
    parser.add_argument("--total_epochs", type=int, default=5)
    parser.add_argument("--val_interval", type=int, default=500)
    parser.add_argument("--save_interval", type=int, default=500)
    parser.add_argument("--lr", type=float, default=1e-5)
    parser.add_argument("--resume_path", type=str, default=None)
    args = parser.parse_args()

    seed_everything(0)

    train_predictor(args)