FedALign/generate_image.py at main · sunnyinAI/FedALign · 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
from copy import deepcopy
import os
from pathlib import Path
import random
import sys
from argparse import Namespace, ArgumentParser
import pickle
import time
from matplotlib import pyplot as plt
import pandas as pd
from typing import Dict, List
import numpy as np
from rich.console import Console
import torch
from torchvision.io import read_image
import torchvision.transforms as transforms

from multiprocessing import cpu_count, Pool
import numpy as np
from pytorch_grad_cam import GradCAM
from pytorch_grad_cam.utils.image import show_cam_on_image
from utils.tools import (
    fix_random_seed,
    update_args_from_dict,
    local_time,
    Logger,
    get_best_device,
)
from model.models import get_model_arch
from torchvision.io import ImageReadMode

# PROJECT_DIR = Path(__file__).parent.parent.parent.absolute()
# sys.path.append(PROJECT_DIR.as_posix())

from data.partition_data import (
    partition_and_statistic,
    get_partition_arguments,
    ALL_DOMAINS,
)
from data.dataset import zero_one_normalize
# from algorithm.server.fedmsfa import FedMSFAServer, get_fedmsfa_argparser
# from algorithm.server.fedms import FedMSServer, get_fedms_argparser
from algorithm.server.fedavg import FedAvgServer, get_fedavg_argparser

# mod = "grad-cam"
mod= "tsne"
assert mod in ["tsne", "augment", "grad-cam"]

# if mod == "augment":
#     server = FedMSServer()
#     server.visualize_augmentation_effect()
# elif mod == "grad-cam":
#     dataset = "pacs"
#     domain = "sketch"
#     model = get_model_arch(model_name="mobile3l")(dataset=dataset)
#     fedavg_checkpoint = f"out/FedAvg/pacs/2024-09-28-15:43:20/{domain}/checkpoint.pth"
#     fedmsfa_checkpoint = f"out/FedMSFA/pacs/num_clients_per_domain_2/AugMix/combine_all/eta_1.0_delta_0.1/{domain}/checkpoint.pth"
#     dataset_path = os.path.join("data", dataset, "raw", domain)
#     labels = os.listdir(dataset_path)
#     for i in range(100):
#         random_label = random.choice(labels)
#         image_path = os.path.join(
#             dataset_path,
#             random_label,
#             random.choice(os.listdir(os.path.join(dataset_path, random_label))),
#         )

#         original_image = read_image(image_path)
#         transform = transforms.Compose(
#             [
#                 (transforms.Resize((224, 224))),
#                 zero_one_normalize(),
#                 # transforms.ToTensor(),
#                 transforms.Normalize(
#                     mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
#                 ),
#             ]
#         )
#         image = transform(original_image)
#         image = image.unsqueeze(0)
#         model = model.cuda()
#         input_image = image.cuda()
#         model.load_state_dict(torch.load(fedavg_checkpoint, weights_only=True)["model"])
#         model.eval()
#         fedavg_prediction = model(input_image).argmax(dim=1).item()

#         model.load_state_dict(
#             torch.load(fedmsfa_checkpoint, weights_only=True)["model"]
#         )
#         model.eval()
#         fedmsfa_prediction = model(input_image).argmax(dim=1).item()
#         label_index = labels.index(random_label)
#         # Check if predictions are different and save the image if they are
#         if fedavg_prediction != fedmsfa_prediction and fedavg_prediction == label_index:
#             print(random_label, fedavg_prediction, fedmsfa_prediction)
#             path2save = os.path.join("image", "grad_cam", local_time())
#             if not os.path.exists(path2save):
#                 os.makedirs(path2save)
#             model.load_state_dict(
#                 torch.load(fedavg_checkpoint, weights_only=True)["model"]
#             )
#             model.eval()
#             target_layer = model.base.features[-1]
#             # Save the original image
#             original_image_path = os.path.join(path2save, "original_image.png")
#             plt.imsave(original_image_path, original_image.permute(1, 2, 0).numpy())

#             cam = GradCAM(model=model, target_layers=[target_layer])
#             grayscale_cam = cam(input_tensor=input_image)
#             image = input_image.squeeze(0).permute(1, 2, 0).cpu().numpy()
#             image = image - image.min()
#             image = image / image.max()
#             cam_image = show_cam_on_image(image, grayscale_cam[0], use_rgb=True)
#             grad_cam_image_path = os.path.join(path2save, "fedavg.png")
#             plt.imsave(grad_cam_image_path, cam_image)

#             model.load_state_dict(
#                 torch.load(fedmsfa_checkpoint, weights_only=True)["model"]
#             )
#             model.eval()
#             target_layer = model.base.features[-1]
#             cam = GradCAM(model=model, target_layers=[target_layer])
#             grayscale_cam = cam(input_tensor=input_image)
#             cam_image = show_cam_on_image(image, grayscale_cam[0], use_rgb=True)
#             plt.imsave(os.path.join(path2save, "fedmsfa.png"), cam_image)
if mod=="tsne":
    args = get_fedavg_argparser().parse_args()
    # path_patition_dir = "2024-09-22-20:17:07"
    path_patition_dir = "2025-01-17-12:03:47"
    # checkpoint_path = (
    #     "out/FedMSFA/pacs/num_clients_per_domain_2/AugMix/combine_all/eta_1.0_delta_0.1"
    # )
    checkpoint_path = (
        "/FedAlign/out/FedAlign/pacs/2025-01-17-12:03:47"
    )
    algo = "FedAlign"
    dataset = "pacs"
    for domain in ALL_DOMAINS[dataset]:
        args.partition_info_dir = os.path.join(path_patition_dir, domain)
        server = FedAvgServer(args=deepcopy(args))
        checkpoint = os.path.join(checkpoint_path, domain, "checkpoint.pth")
        # print(type(checkpoint))
        # print(checkpoint)

        server.resume_checkpoint(checkpoint)
        server.draw_feature_distribution(algo=algo)