KeyError: 'mask' about clean-pvnet

[pengsida]

I have fixed the bug.
You could directly test the evaluation code using python train_net.py --test --cfg_file configs/tless/tless_01.yaml.

[pengsida]

What is your command?
I can train the network with python train_net.py --cfg_file configs/tless/tless_01.yaml model tmp.

[firstdeep]

What is your command?
I can train the network with python train_net.py --cfg_file configs/tless/tless_01.yaml model tmp.

I have the same issue.

My command is = python train_net.py --cfg_file configs/tless/tless_01.yaml model tless_models/tless_01 eval_ep 1

There seems to be a problem with reading data in the area of evaluation.
https://github.com/zju3dv/clean-pvnet/blob/master/train_net.py#L35

[pengsida]

I fixed the bug.

[firstdeep]

@pengsida

Thanks for fix the bug.

I have a question.
The test data of tless consists of 504 images. (test_primesense = 06)
But I wonder why there are 1,008 images that are actually loaded.

Thanks for reply

[pengsida]

I evaluate every objects on the whole dataset.
1,008 is the number of images containing a specific object in the dataset.

[Ji-Hu]

@pengsida

Thank you for sharing your great achievements.

I'd like to ask you a few questions.

#1.
https://github.com/zju3dv/clean-pvnet/blob/master/lib/evaluators/tless_test/pvnet.py#L43
In case of ADD, I think it will be judged that it is correct if it is within 10% of the diameter. Why is there a reason to divide the diameter by 1000?

I also want to ask about the difference between adi_metrix and add_metrix.

#2.
I wonder if there is a reason for the evaluation of the input image by cropping only the test_object part (input size = 256 * 256)

I think we should put the original test image(image size = 540 * 720 / Not cropping) in the input.

[pengsida]

1. The unit of the object diameter is mm.
2. adi is a variant of add for symmetric objects.
3. To handle multiple instances.

[kashyap92]

Hello @pengsida ,
Thank you for your response. 'KeyError: 'mask' is now resolved'.
However, after 4 epoch i am facing new error now.

My training command is : python train_net.py --cfg_file configs/tless/tless_01.yaml

Error:
Traceback (most recent call last):
File "train_net.py", line 57, in
main()
File "train_net.py", line 53, in main
train(cfg, network)
File "train_net.py", line 35, in train
trainer.val(epoch, val_loader, evaluator, recorder)
File "/datasets/Mahesh_kashyap/PoseEst/pvnet-clean/clean-pvnet/lib/train/trainers/trainer.py", line 82, in val
output, loss, loss_stats, image_stats = self.network(batch)
File "/home/belakavadi/anaconda3/envs/pvnet/lib/python3.7/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, kwargs)
File "/home/belakavadi/anaconda3/envs/pvnet/lib/python3.7/site-packages/torch/nn/parallel/data_parallel.py", line 153, in forward
return self.gather(outputs, self.output_device)
File "/home/belakavadi/anaconda3/envs/pvnet/lib/python3.7/site-packages/torch/nn/parallel/data_parallel.py", line 165, in gather
return gather(outputs, output_device, dim=self.dim)
File "/home/belakavadi/anaconda3/envs/pvnet/lib/python3.7/site-packages/torch/nn/parallel/scatter_gather.py", line 67, in gather
return gather_map(outputs)
File "/home/belakavadi/anaconda3/envs/pvnet/lib/python3.7/site-packages/torch/nn/parallel/scatter_gather.py", line 62, in gather_map
return type(out)(map(gather_map, zip(outputs)))
File "/home/belakavadi/anaconda3/envs/pvnet/lib/python3.7/site-packages/torch/nn/parallel/scatter_gather.py", line 62, in gather_map
return type(out)(map(gather_map, zip(outputs)))
TypeError: zip argument #1 must support iteration

Could you please help me resolved this?

Regards,
Mahesh kashyap.