I inspect your attached mask label png file but find that it has pixel value as below.
[ 0 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 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
145 146 147 148 149 150 151 152 153 155 156 157 158 159 163 164 166 167
168 169 170 171 172 173 174 176 177 178 179 180 181 182 183 184 185 186
187 188 189 190 191 192 193 195 196 197 198 199 200 201 202 203 204 205
206 207 208 209 210 220 238]
It does not match your training classes.
Please note that the pixel integer value should be equal to the value of the label_id provided in the spec.
UNet expects the images and corresponding masks encoded as images. Each mask image is a single-channel image, where every pixel is assigned an integer value that represents the segmentation class.
See Data Annotation Format - NVIDIA Docs
More,
See UNET - NVIDIA Docs
Dice loss for multi-class segmentation is not supported.
So for multiple classes, please use “cross_entropy”.
Also, if want to get the highest accuracy, suggest to use the powerful backbone: vanilla_unet_dynamic