adaPT/main.py at main · ispamm/adaPT · 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
import hydra
from omegaconf import DictConfig
import torch
import wandb
import pytorch_lightning as pl
from pytorch_lightning.loggers import WandbLogger
from datasets.ModelNet40Ply2048 import ModelNet40Ply2048DataModule
from model import Adapt_classf_pl
import os
import numpy as np
import random
import matplotlib.pyplot as plt
from pytorch_lightning.callbacks import LearningRateMonitor
from point_transformer_cls import PCT_PL
from fvcore.nn import FlopCountAnalysis

os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "3"
os.environ["WANDB_API_KEY"] = "04a5d6fba030b76e5b620f5bd6509cf7dffebb8b"

def train(cfg, train_loader, test_loader):

    device = "cuda" if cfg.cuda else "cpu"
    if cfg.model.name == "Adapt_classf":
        model = Adapt_classf_pl(cfg, cfg.model.embed_dim, cfg.n_points, cfg.n_classes, cfg.model.n_blocks, cfg.model.groups)
    elif cfg.model.name == "PCT_reproduce":
        model = PCT_PL()
    else:
        raise Exception("Model not supported")

    if cfg.wandb:
        wandb_logger = WandbLogger(name=cfg.experiment.name, project=cfg.experiment.project)
        wandb_logger.watch(model)
        wandb_logger.log_hyperparams(cfg)
        wandb_logger.log_hyperparams(model.hparams)
        lr_monitor = LearningRateMonitor(logging_interval='step')

    trainer = pl.Trainer(max_epochs=cfg.train.epochs, accelerator=device, logger=[wandb_logger] if cfg.wandb else None, devices=1, gradient_clip_val=2, callbacks=[lr_monitor] if cfg.wandb else None)#, default_root_dir='saved_models')
    trainer.fit(model, train_loader, test_loader)

    if cfg.wandb:
        wandb.finish()

    return None

def test(cfg, test_loader):
    raise NotImplementedError

def visualize(cfg):
    raise NotImplementedError

def eval_time(cfg,x):

    device = "cuda" if cfg.cuda else "cpu"
    x = torch.randn(cfg.train.batch_size, 512, 3)
    x = x.to(device)
    model = Adapt_classf_pl(cfg, cfg.model.embed_dim, cfg.n_points, cfg.n_classes, cfg.model.n_blocks, cfg.model.groups)
    model = model.to(device)
    model.eval()
    flops = FlopCountAnalysis(model, x)
    print(flops.total()/cfg.train.batch_size)


@hydra.main(config_path=".", config_name="config", version_base=None)
def main(cfg: DictConfig):

    torch.set_float32_matmul_precision('high')
    pl.seed_everything(cfg.experiment.seed)
    if cfg.wandb:
        wandb.login()
        wandb.init(config=cfg)

    cfg.cuda = cfg.cuda and torch.cuda.is_available()
    if cfg.cuda:
        print(
            'Using GPU : ' + str(torch.cuda.current_device()) + ' from ' + str(torch.cuda.device_count()) + ' devices')
        torch.cuda.manual_seed(cfg.experiment.seed)
    else:
        print('Using CPU')

    if cfg.experiment.dataset == "ModelNet40":
        dataset = ModelNet40Ply2048DataModule(batch_size=cfg.train.batch_size)
    else:
        raise Exception("Dataset not supported")
    dataset.setup()
    train_loader = dataset.train_dataloader()
    test_loader = dataset.val_dataloader()
    cfg.n_classes = dataset.num_classes
    cfg.n_points = dataset.num_points

    if not cfg.eval:
        #eval_time(cfg, next(iter(test_loader))[0][:,:,:])
        train(cfg, train_loader, test_loader)
    else:
        if not cfg.visualize_pc:
            test(cfg, test_loader)
        else:
            visualize(cfg)

if __name__ == "__main__":
    main()


################################################## OLD CODE ##################################################
"""
    with torch.no_grad():
        data, label = next(iter(test_loader))
        decisions = []
        model = model.to(device)
        data = data.to(device)
        for budg in range(cfg.train.n_budgets):
            _, decision = model(data, budg=budg)
            decisions.append(decision[-1].reshape(-1).cpu())

        print(decisions[0].shape)
        decisions = torch.stack(decisions, dim=0).sum(dim=0)
        print(decisions.shape)
        optimal = torch.zeros_like(decisions)
        targets = cfg.model.drop_rate[-1]*(torch.arange(cfg.train.n_budgets)/(cfg.train.n_budgets-1))
        random_decision = torch.zeros_like(decisions)
        for targ in targets:
            optimal[:int(targ*len(optimal))] += 1
            ind = random.sample(range(len(optimal)), int(targ*len(optimal)))
            random_decision[ind] += 1

        print(targets)
        print(decisions)
        print(optimal)
        optimal = optimal.cpu().numpy()
        decisions = decisions.cpu().numpy()
        optimal_histo = np.histogram(optimal, bins=cfg.train.n_budgets+1)
        decision_histo = np.histogram(decisions, bins=cfg.train.n_budgets+1)
        random_decision_histo = np.histogram(random_decision, bins=cfg.train.n_budgets+1)

        fig = plt.figure()
        plt.hist([optimal, decisions, random_decision], alpha=0.5, bins=cfg.train.n_budgets+1, label=["optimal", "decision", "random_decision"])
        plt.legend(loc='upper right')
        plt.savefig("histo.png")
"""