Training Problem about cbst HOT 9 CLOSED

It seems the error is resulted by the different resolutions of GTA5 images. In most cases, the resolution of GTA5 is 1052 x 1914. But there are some images that have different resolution, e.g. 1046 x 1914. So I suggest you to resize all the GTA5 images to the same resolutions 1052 x 1914. That might solve the problem.

from cbst.

I've found it, but still thank you very much.

from cbst.

I run the synthia2Cityscapes self-training code,and meet the similar question:

Do you know how to solve it?

from cbst.

I run the synthia2Cityscapes self-training code,and meet the similar question:

Do you know how to solve it?
When loading image of synthia dataset, you have to pay attention to the numbers of channel of the loaded image, only 1 of the 3 channels is useful or what we want. I'm not sure this is the point, but you can check the code to see if this is the reason casued the error.

from cbst.

I run the synthia2Cityscapes self-training code,and meet the similar question:

Do you know how to solve it?
When loading image of synthia dataset, you have to pay attention to the numbers of channel of the loaded image, only 1 of the 3 channels is useful or what we want. I'm not sure this is the point, but you can check the code to see if this is the reason casued the error.

Just as Jiale suggested, check the dimension of your label map to see if they have 1 or 3 channels. I think that's the reason causing this problem and you may consider generating label map with 1 channel.

from cbst.

I run the synthia2Cityscapes self-training code,and meet the similar question:

Do you know how to solve it?
When loading image of synthia dataset, you have to pay attention to the numbers of channel of the loaded image, only 1 of the 3 channels is useful or what we want. I'm not sure this is the point, but you can check the code to see if this is the reason casued the error.

Just as Jiale suggested, check the dimension of your label map to see if they have 1 or 3 channels. I think that's the reason causing this problem and you may consider generating label map with 1 channel.
I download tha synthia dataset from the url you provided, so shall I need process the labels from dataset so as to become one channel images?
Following is my input training commond:
python issegm/solve_AO.py --num-round 6 --test-scales 1850 --scale-rate-range 0.7,1.3 --dataset synthia --dataset-tgt cityscapes --split train --split-tgt train --data-root /home/qiu/下载/CBST/data/RAND_CITYSCAPES --data-root-tgt /home/qiu/下载/CBST/data/cityscapes --output syn2city/cbst --model cityscapes_rna-a1_cls16_s8 --weights models/synthia_rna-a1_cls16_s8_ep-0000.params --batch-images 2 --crop-size 500 --origin-size 1280 --origin-size-tgt 2048 --init-tgt-port 0.2 --init-src-port 0.02 --max-src-port 0.06 --seed-int 0 --mine-port 0.8 --mine-id-number 3 --mine-thresh 0.001 --base-lr 1e-4 --to-epoch 2 --source-sample-policy cumulative --self-training-script issegm/solve_ST.py --kc-policy global --prefetch-threads 2 --gpus 0 --with-prior False

from cbst.

I run the synthia2Cityscapes self-training code,and meet the similar question:

Do you know how to solve it?
When loading image of synthia dataset, you have to pay attention to the numbers of channel of the loaded image, only 1 of the 3 channels is useful or what we want. I'm not sure this is the point, but you can check the code to see if this is the reason casued the error.

Just as Jiale suggested, check the dimension of your label map to see if they have 1 or 3 channels. I think that's the reason causing this problem and you may consider generating label map with 1 channel.
I download tha synthia dataset from the url you provided, so shall I need process the labels from dataset so as to become one channel images?
Following is my input training commond:
python issegm/solve_AO.py --num-round 6 --test-scales 1850 --scale-rate-range 0.7,1.3 --dataset synthia --dataset-tgt cityscapes --split train --split-tgt train --data-root /home/qiu/下载/CBST/data/RAND_CITYSCAPES --data-root-tgt /home/qiu/下载/CBST/data/cityscapes --output syn2city/cbst --model cityscapes_rna-a1_cls16_s8 --weights models/synthia_rna-a1_cls16_s8_ep-0000.params --batch-images 2 --crop-size 500 --origin-size 1280 --origin-size-tgt 2048 --init-tgt-port 0.2 --init-src-port 0.02 --max-src-port 0.06 --seed-int 0 --mine-port 0.8 --mine-id-number 3 --mine-thresh 0.001 --base-lr 1e-4 --to-epoch 2 --source-sample-policy cumulative --self-training-script issegm/solve_ST.py --kc-policy global --prefetch-threads 2 --gpus 0 --with-prior False

And did you process the labels before training? I didn't change anything else ,or maybe is my input command wrong?

from cbst.

I run the synthia2Cityscapes self-training code,and meet the similar question:

Do you know how to solve it?
When loading image of synthia dataset, you have to pay attention to the numbers of channel of the loaded image, only 1 of the 3 channels is useful or what we want. I'm not sure this is the point, but you can check the code to see if this is the reason casued the error.

Just as Jiale suggested, check the dimension of your label map to see if they have 1 or 3 channels. I think that's the reason causing this problem and you may consider generating label map with 1 channel.
I download tha synthia dataset from the url you provided, so shall I need process the labels from dataset so as to become one channel images?
Following is my input training commond:
python issegm/solve_AO.py --num-round 6 --test-scales 1850 --scale-rate-range 0.7,1.3 --dataset synthia --dataset-tgt cityscapes --split train --split-tgt train --data-root /home/qiu/下载/CBST/data/RAND_CITYSCAPES --data-root-tgt /home/qiu/下载/CBST/data/cityscapes --output syn2city/cbst --model cityscapes_rna-a1_cls16_s8 --weights models/synthia_rna-a1_cls16_s8_ep-0000.params --batch-images 2 --crop-size 500 --origin-size 1280 --origin-size-tgt 2048 --init-tgt-port 0.2 --init-src-port 0.02 --max-src-port 0.06 --seed-int 0 --mine-port 0.8 --mine-id-number 3 --mine-thresh 0.001 --base-lr 1e-4 --to-epoch 2 --source-sample-policy cumulative --self-training-script issegm/solve_ST.py --kc-policy global --prefetch-threads 2 --gpus 0 --with-prior False

And did you process the labels before training? I didn't change anything else ,or maybe is my input command wrong?

I just checked the original dataset, labels have only one channel. The problem may be caused by different version of PIL image loading function. (In my environment it works well). If the image loading function of PIL.Image loads image with 3 channels, just use one of them.

from cbst.

I run the synthia2Cityscapes self-training code,and meet the similar question:

Do you know how to solve it?
When loading image of synthia dataset, you have to pay attention to the numbers of channel of the loaded image, only 1 of the 3 channels is useful or what we want. I'm not sure this is the point, but you can check the code to see if this is the reason casued the error.

Just as Jiale suggested, check the dimension of your label map to see if they have 1 or 3 channels. I think that's the reason causing this problem and you may consider generating label map with 1 channel.
I download tha synthia dataset from the url you provided, so shall I need process the labels from dataset so as to become one channel images?
Following is my input training commond:
python issegm/solve_AO.py --num-round 6 --test-scales 1850 --scale-rate-range 0.7,1.3 --dataset synthia --dataset-tgt cityscapes --split train --split-tgt train --data-root /home/qiu/下载/CBST/data/RAND_CITYSCAPES --data-root-tgt /home/qiu/下载/CBST/data/cityscapes --output syn2city/cbst --model cityscapes_rna-a1_cls16_s8 --weights models/synthia_rna-a1_cls16_s8_ep-0000.params --batch-images 2 --crop-size 500 --origin-size 1280 --origin-size-tgt 2048 --init-tgt-port 0.2 --init-src-port 0.02 --max-src-port 0.06 --seed-int 0 --mine-port 0.8 --mine-id-number 3 --mine-thresh 0.001 --base-lr 1e-4 --to-epoch 2 --source-sample-policy cumulative --self-training-script issegm/solve_ST.py --kc-policy global --prefetch-threads 2 --gpus 0 --with-prior False

And did you process the labels before training? I didn't change anything else ,or maybe is my input command wrong?

I just checked the original dataset, labels have only one channel. The problem may be caused by different version of PIL image loading function. (In my environment it works well). If the image loading function of PIL.Image loads image with 3 channels, just use one of them.

Thank you ,it works.

from cbst.