ajpa2024/layer_cams.py at main · vkola-lab/ajpa2024 · 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
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
import torch
from torch_geometric.loader import DataLoader
import torch_geometric.transforms as T
import torch.nn.functional as F

import os
import random
from tqdm import tqdm
import argparse
import numpy as np
import matplotlib.pyplot as plt

from PIL import Image
import PIL
PIL.Image.MAX_IMAGE_PIXELS = 933120000

from gnn import GNN
from visualize import generate_relevance, plot_heat_maps
from evaluate import grad_cam
from util import read_file, find_dataset_using_name

def generate_cams(args, model, device, multiple_loaders, index=None):

    model.eval()

    y_true = []
    y_pred = []

    os.makedirs(args.output_folder, exist_ok=True)

    for loader in multiple_loaders:

        true_labels = list(loader.dataset.classdict.keys())
        to_be_predicted_classes = list(loader.dataset.to_be_predicted_classes.keys())

        for step, graph in enumerate(tqdm(loader, desc="Iteration")):

            graph = graph.to(device)
            slide_name = graph.slide_path[0]
            print(slide_name)
            # print(graph.node_coords)

            if graph.x.shape[0] == 1:
                pass
            else:

                # GENERATE VISUALIZATION :
                transformer_attribution, output, y_pred = generate_relevance(model, graph, index=index)
                if index is not None:
                    y_pred = index

                print("logits: ", output)

                prob = F.softmax(output, dim=1)
                prob = prob.squeeze()

                print("Slide: {}, True Class: {}, Predicted Class: {}(p={:.3f})".format(slide_name, true_labels[graph.y], to_be_predicted_classes[y_pred], prob[y_pred].item()))
                del output

                slide_root = os.path.join('/SeaExp/Rushin/datasets/', args.dataset_name.upper(), 'WSIs')
                plt = plot_heat_maps(graph, scores=transformer_attribution, prob=prob[index], slide_root=slide_root, clamp=0.05, save_path=args.output_folder, overlay=True)
                # Use numpy to save attention_blend image to a file
                # attention_blend = Image.fromarray(attention_blend)
                # attention_blend.save(os.path.join(args.output_folder, "{}_{}(fold{})_cam.png".format(slide_name, to_be_predicted_classes[y_pred], args.fold_idx)))

                plt.savefig(os.path.join(args.output_folder, "{}_{}(fold{})_cam.png".format(slide_name, to_be_predicted_classes[y_pred], args.fold_idx)))
                plt.close()


def main():
    # Training settings
    parser = argparse.ArgumentParser(description='GNN baselines on ogbg-ppa data with Pytorch Geometrics')
    parser.add_argument('--device', type=int, default=0,
                        help='which gpu to use if any (default: 0)')

    parser.add_argument('--gnn', type=str, default='gin',
                        help='GNN gin, gin-virtual, or gcn, or gcn-virtual (default: gin-virtual)')
    parser.add_argument('--num_layer', type=int, default=3,
                        help='number of GNN message passing layers (default: 5)')
    parser.add_argument('--emb_dim', type=int, default=128,
                        help='dimensionality of hidden units in GNNs (default: 300)')
    parser.add_argument('--drop_ratio', type=float, default=0,
                        help='dropout ratio (default: 0.5)')
    parser.add_argument('--jk', type=str, default='sum',
                        help='Jumping knowledge aggregations : last | sum')
    parser.add_argument('--graph_pooling', type=str, default='gmt',
                        help='Graph pooling type : sum | mean | max | attention | set2set')

    parser.add_argument('--seed', type=int, default=42,
                        help='random seed for splitting the dataset into 10 (default: 0)')
    parser.add_argument('--batch_size', type=int, default=1,
                        help='input batch size for training (default: 1)')
    parser.add_argument('--num_workers', type=int, default=4,
                        help='number of workers (default: 0)')

    parser.add_argument('--dataset', nargs='+', type=str, default="cptac",
                        help='dataset name (default: pcga | tcga | cis )')
    parser.add_argument('--phase', type=str, default="cams",
                        help='dataset phase : train | test | cams')
    parser.add_argument('--n_classes', type=int, default=3,
                        help='number of classes')
    parser.add_argument('--data_config', type=str, default="ctranspath_files",
                        help='dataset config i.e tile size and bkg content (default: simclr_files)')
    parser.add_argument('--fdim', type=int, default=768,
                        help='expected feature dim for each node.')
    parser.add_argument('--patch_size', type=int, default=256,
                        help='patch_size')

    parser.add_argument('-L', '--ignore-bounds', dest='limit_bounds',
		default=True, action='store_false', help='display entire scan area')


    parser.add_argument('--fold_idx', type=int, default=0,
                        help='The fold to consider.')
    parser.add_argument('--run_name', type=str, default="easy-wind-35",
                        help='run name to get all model logs')

    parser.add_argument('--output', type = str, default = "logs", help='Folder in which to save the tsne plots')
    args = parser.parse_args()

    ### set up seeds and gpu device
    torch.manual_seed(args.seed)
    torch.cuda.manual_seed(args.seed)
    torch.cuda.manual_seed_all(args.seed)
    np.random.seed(args.seed)
    random.seed(args.seed)
    torch.backends.cudnn.benchmark = False
    torch.backends.cudnn.deterministic = True

    device = torch.device("cuda:" + str(args.device)) if torch.cuda.is_available() else torch.device("cpu")

    args.slide_feats_folder = {}

    fold_idx = args.fold_idx
    test_loaders = []

    for item in args.dataset:

        args.dataset_name = item

        dataset_class = find_dataset_using_name(item)
        print(dataset_class)

        ### automatic dataloading and splitting
        root = os.path.join('/SeaExp/Rushin/datasets', item.upper(), args.data_config)
        wsi_file = os.path.join('/SeaExp/Rushin/datasets', item.upper(), '%s_%s.txt' % (item.upper(), args.phase))
        wsi_ids = read_file(wsi_file)

        dataset = dataset_class(root, wsi_ids, args.fdim, n_classes=args.n_classes, isTrain=False, transform=T.ToSparseTensor(remove_edge_index=False))
        test_loaders.append(DataLoader(dataset, batch_size=1, shuffle=False, num_workers=args.num_workers))

    log_path = os.path.join('logs', "{}_fold_{}".format(args.run_name, fold_idx))

    model = GNN(gnn_type = 'gin', num_class = dataset.num_classes, num_layer = args.num_layer, input_dim = args.fdim, emb_dim = args.emb_dim, drop_ratio = args.drop_ratio, JK = args.jk, graph_pooling = args.graph_pooling).to(device)

    model.load_state_dict(torch.load(os.path.join(log_path, "final_model_{}_fold_{}.pth".format(args.run_name, fold_idx))))
    model = model.to(device)
    print("model weights loaded successfully")

    args.output_folder = os.path.join(args.output, args.run_name+"_{}_{}".format(args.patch_size, args.phase), args.dataset_name)

    for i in range(0, args.n_classes):
        args.index = i
        print("Generating CAMs for class {}".format(i))
        generate_cams(args, model, device, test_loaders, args.index)

if __name__ == "__main__":
    main()