agcompvis/dataset.py at main · zmswanson/agcompvis · 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
import torch
from torch import tensor
from torch.nn.functional import normalize
from torchvision import transforms
import torchvision
from torchvision.io import read_image
from torchvision.transforms import ToTensor
from torch.utils import data
import numpy as np
import os
from PIL import Image

class AgVisionDataSet(data.Dataset):
    def __init__(self, select_dataset="train", transform=None, start_index=0):
        if select_dataset in ['train', 'val', 'test']:
            self.dataset = select_dataset
        else:
            raise ValueError(f"{select_dataset} is not a valid dataset. Valid datasets include \"train\", \"val\", and \"test\".")

        rgb_path = f"dataset/{self.dataset}/images/rgb/"
        nir_path = f"dataset/{self.dataset}/images/nir/"

        self.rgb_inputs = [rgb_path + s for s in sorted(os.listdir(rgb_path))[start_index:]]
        self.nir_inputs = [nir_path + s for s in sorted(os.listdir(nir_path))[start_index:]]

        if self.dataset in ["train", "val"]:
            dp1_path = f"dataset/{self.dataset}/labels/double_plant/"
            dd2_path = f"dataset/{self.dataset}/labels/drydown/"
            er3_path = f"dataset/{self.dataset}/labels/endrow/"
            nd4_path = f"dataset/{self.dataset}/labels/nutrient_deficiency/"
            ps5_path = f"dataset/{self.dataset}/labels/planter_skip/"
            wa6_path = f"dataset/{self.dataset}/labels/water/"
            ww7_path = f"dataset/{self.dataset}/labels/waterway/"
            wc8_path = f"dataset/{self.dataset}/labels/weed_cluster/"

            self.double_plant = [dp1_path + s for s in sorted(os.listdir(dp1_path))[start_index:]]
            self.drydown      = [dd2_path + s for s in sorted(os.listdir(dd2_path))[start_index:]]
            self.endrow       = [er3_path + s for s in sorted(os.listdir(er3_path))[start_index:]]
            self.nutrient_def = [nd4_path + s for s in sorted(os.listdir(nd4_path))[start_index:]]
            self.planter_skip = [ps5_path + s for s in sorted(os.listdir(ps5_path))[start_index:]]
            self.water        = [wa6_path + s for s in sorted(os.listdir(wa6_path))[start_index:]]
            self.waterway     = [ww7_path + s for s in sorted(os.listdir(ww7_path))[start_index:]]
            self.weed_cluster = [wc8_path + s for s in sorted(os.listdir(wc8_path))[start_index:]]

        bound_path = f"dataset/{self.dataset}/boundaries/"
        self.boundaries = [bound_path + s for s in sorted(os.listdir(bound_path))[start_index:]]

        self.transform = transform
        self.inputs_dtype = torch.float32
        self.targets_dtype = torch.float32

    def __len__(self):
        return len(self.rgb_inputs)

    def __getitem__(self, index: int):
        # Select the sample
        rgb_input_ID = self.rgb_inputs[index]
        nir_input_ID = self.nir_inputs[index]
        boundary_ID  = self.boundaries[index]

        if self.dataset in ["train", "val"]:
            double_plant_ID = self.double_plant[index]
            drydown_ID      = self.drydown[index]
            endrow_ID       = self.endrow[index]
            nutrient_def_ID = self.nutrient_def[index]
            planter_skip_ID = self.planter_skip[index]
            water_ID        = self.water[index]
            waterway_ID     = self.waterway[index]
            weed_cluster_ID = self.weed_cluster[index]

        # Load input
        x0 = read_image(rgb_input_ID).type(self.inputs_dtype)
        x1 = read_image(nir_input_ID).type(self.inputs_dtype)
        x = normalize(torch.cat([x0, x1], dim=0)) # combine the nir and rgb data

        y = torch.zeros((1, 512, 512))

        if self.dataset in ["train", "val"]:
            y = torch.cat([y, read_image(double_plant_ID).type(self.targets_dtype) / 255], dim=0)
            y = torch.cat([y, read_image(drydown_ID).type(self.targets_dtype) * 2 / 255], dim=0)
            y = torch.cat([y, read_image(endrow_ID).type(self.targets_dtype) * 3 / 255], dim=0)
            y = torch.cat([y, read_image(nutrient_def_ID).type(self.targets_dtype) * 4 / 255], dim=0)
            y = torch.cat([y, read_image(planter_skip_ID).type(self.targets_dtype) * 5 / 255], dim=0)
            y = torch.cat([y, read_image(water_ID).type(self.targets_dtype) * 6 / 255], dim=0)
            y = torch.cat([y, read_image(waterway_ID).type(self.targets_dtype) * 7 / 255], dim=0)
            y = torch.cat([y, read_image(weed_cluster_ID).type(self.targets_dtype) * 8 / 255], dim=0)
            y = torch.argmax(y, dim=0, keepdim=True)

            y = torch.squeeze(y)

        # Preprocessing
        if self.transform is not None:
            x, y = self.transform(x, y)


        b = read_image(boundary_ID).type(self.inputs_dtype) / 255.0
        b = b[0, :, :]

        if self.dataset in ['test']:
            id = os.path.basename(rgb_input_ID).split('.')[0]
            return x, y, b, id

        b = torch.reshape(b, (1, 512, 512))
        b = torch.repeat_interleave(b, 9, dim=0)
        return x, y, b

if __name__ == '__main__':
    training_dataset = AgVisionDataSet(select_dataset='train', transform=None)

    training_dataloader = data.DataLoader(dataset=training_dataset, batch_size=2, shuffle=True)

    x, y = next(iter(training_dataloader))

    print(f'x = shape: {x.shape}; type: {x.dtype}')
    print(f'x = min: {x.min()}; max: {x.max()}')
    print(f'y = shape: {y.shape}; class: {y.unique()}; type: {y.dtype}')

    mapping = {'bg': (0, 0, 0),
               'dp': (255, 0, 0),
               'dd': (255, 128, 0),
               'er': (255, 255, 0),
               'nd': (128, 255, 0),
               'ps': (0, 255, 0),
               'wa': (255, 0, 255),
               'ww': (0, 0, 255),
               'wc': (128, 0, 255)
    }

    print(y[1])
    print(y[1].shape)
    z = torch.squeeze(y[1])

    seg_img = np.zeros((512,512,3))
    seg_img[z == 1] = mapping['dp']
    seg_img[z == 2] = mapping['dd']
    seg_img[z == 3] = mapping['er']
    seg_img[z == 4] = mapping['nd']
    seg_img[z == 5] = mapping['ps']
    seg_img[z == 6] = mapping['wa']
    seg_img[z == 7] = mapping['ww']
    seg_img[z == 8] = mapping['wc']

    img = Image.fromarray(seg_img, 'RGB')
    img.show()