xwhkkk
commented
11 months ago When will the code for semantic segmentation be publicly available? I cannot reproduce the miou=75 in your paper myself. I don't know where I went wrong. Here are my training results
+---------------+-------+-------+--------+-----------+--------+-------+ | Class | IoU | Acc | Fscore | Precision | Recall | Dice | +---------------+-------+-------+--------+-----------+--------+-------+ | road | 94.76 | 98.47 | 97.31 | 96.18 | 98.47 | 97.31 | | sidewalk | 65.0 | 74.29 | 78.79 | 83.87 | 74.29 | 78.79 | | building | 83.27 | 91.41 | 90.87 | 90.33 | 91.41 | 90.87 | | wall | 28.94 | 39.38 | 44.89 | 52.21 | 39.38 | 44.89 | | fence | 25.94 | 33.52 | 41.19 | 53.42 | 33.52 | 41.19 | | pole | 33.56 | 41.68 | 50.25 | 63.25 | 41.68 | 50.25 | | traffic light | 29.34 | 36.49 | 45.36 | 59.95 | 36.49 | 45.36 | | traffic sign | 40.46 | 47.41 | 57.61 | 73.41 | 47.41 | 57.61 | | vegetation | 86.34 | 94.39 | 92.67 | 91.01 | 94.39 | 92.67 | | terrain | 48.5 | 54.13 | 65.32 | 82.33 | 54.13 | 65.32 | | sky | 91.27 | 95.62 | 95.43 | 95.25 | 95.62 | 95.43 | | person | 50.87 | 76.7 | 67.43 | 60.16 | 76.7 | 67.43 | | rider | 20.13 | 27.86 | 33.51 | 42.02 | 27.86 | 33.51 | | car | 82.34 | 92.72 | 90.32 | 88.03 | 92.72 | 90.32 | | truck | 31.04 | 36.14 | 47.37 | 68.74 | 36.14 | 47.37 | | bus | 40.0 | 55.51 | 57.14 | 58.87 | 55.51 | 57.14 | | train | 26.41 | 35.33 | 41.79 | 51.14 | 35.33 | 41.79 | | motorcycle | 9.37 | 10.13 | 17.13 | 55.36 | 10.13 | 17.13 | | bicycle | 48.84 | 66.74 | 65.63 | 64.56 | 66.74 | 65.63 | +---------------+-------+-------+--------+-----------+--------+-------+ 2023-10-11 12:43:28,826 - mmseg - INFO - Summary: 2023-10-11 12:43:28,827 - mmseg - INFO - +-------+-------+-------+---------+------------+---------+-------+ | aAcc | mIoU | mAcc | mFscore | mPrecision | mRecall | mDice | +-------+-------+-------+---------+------------+---------+-------+ | 90.65 | 49.28 | 58.31 | 62.11 | 70.0 | 58.31 | 62.11 | +-------+-------+-------+---------+------------+---------+-------+
Hello, how do you train the segmentation model with MMsegmentation? I can not reproduce the mIou either.
1377534928
When will the code for semantic segmentation be publicly available? I cannot reproduce the miou=75 in your paper myself. I don't know where I went wrong. Here are my training results
+---------------+-------+-------+--------+-----------+--------+-------+ | Class | IoU | Acc | Fscore | Precision | Recall | Dice | +---------------+-------+-------+--------+-----------+--------+-------+ | road | 94.76 | 98.47 | 97.31 | 96.18 | 98.47 | 97.31 | | sidewalk | 65.0 | 74.29 | 78.79 | 83.87 | 74.29 | 78.79 | | building | 83.27 | 91.41 | 90.87 | 90.33 | 91.41 | 90.87 | | wall | 28.94 | 39.38 | 44.89 | 52.21 | 39.38 | 44.89 | | fence | 25.94 | 33.52 | 41.19 | 53.42 | 33.52 | 41.19 | | pole | 33.56 | 41.68 | 50.25 | 63.25 | 41.68 | 50.25 | | traffic light | 29.34 | 36.49 | 45.36 | 59.95 | 36.49 | 45.36 | | traffic sign | 40.46 | 47.41 | 57.61 | 73.41 | 47.41 | 57.61 | | vegetation | 86.34 | 94.39 | 92.67 | 91.01 | 94.39 | 92.67 | | terrain | 48.5 | 54.13 | 65.32 | 82.33 | 54.13 | 65.32 | | sky | 91.27 | 95.62 | 95.43 | 95.25 | 95.62 | 95.43 | | person | 50.87 | 76.7 | 67.43 | 60.16 | 76.7 | 67.43 | | rider | 20.13 | 27.86 | 33.51 | 42.02 | 27.86 | 33.51 | | car | 82.34 | 92.72 | 90.32 | 88.03 | 92.72 | 90.32 | | truck | 31.04 | 36.14 | 47.37 | 68.74 | 36.14 | 47.37 | | bus | 40.0 | 55.51 | 57.14 | 58.87 | 55.51 | 57.14 | | train | 26.41 | 35.33 | 41.79 | 51.14 | 35.33 | 41.79 | | motorcycle | 9.37 | 10.13 | 17.13 | 55.36 | 10.13 | 17.13 | | bicycle | 48.84 | 66.74 | 65.63 | 64.56 | 66.74 | 65.63 | +---------------+-------+-------+--------+-----------+--------+-------+ 2023-10-11 12:43:28,826 - mmseg - INFO - Summary: 2023-10-11 12:43:28,827 - mmseg - INFO - +-------+-------+-------+---------+------------+---------+-------+ | aAcc | mIoU | mAcc | mFscore | mPrecision | mRecall | mDice | +-------+-------+-------+---------+------------+---------+-------+ | 90.65 | 49.28 | 58.31 | 62.11 | 70.0 | 58.31 | 62.11 | +-------+-------+-------+---------+------------+---------+-------+