RuntimeError: The shape of the mask [1, 1, 128, 128] at index 1 does not match the shape of the indexed tensor [1, 80, 128, 128] at index 1

[li10141110]

Hi, thank you for your great work. I build a coco style object detection dataset with only one category. According to your project, I rectified CenterNet-104.json file and finally, I was stuck by the following error. I know that the shape of the mask is [1, 1, 128, 128] (my batch_size==1), I don't know how to change it from [1, 80, 128, 128] to [1, 1, 128, 128] accordingly. Any help would be appreciated!

###############################################
/home/imdl/anaconda3/envs/centernet/bin/python /home/imdl/workspace/CenterNet-master_train_custom_dataset/train.py CenterNet-104
loading all datasets...
using 1 threads
loading from cache file: cache/coco_trainval2014.pkl
loading annotations into memory...
Done (t=0.00s)
creating index...
index created!
loading from cache file: cache/coco_minival2014.pkl
loading annotations into memory...
Done (t=0.00s)
creating index...
index created!
system config...
{'batch_size': 1,
'cache_dir': 'cache',
'chunk_sizes': [1],
'config_dir': 'config',
'data_dir': '../data',
'data_rng': <mtrand.RandomState object at 0x7f7bf9c02870>,
'dataset': 'MSCOCO',
'decay_rate': 10,
'display': 5,
'learning_rate': 0.00025,
'max_iter': 100000,
'nnet_rng': <mtrand.RandomState object at 0x7f7bf9c028b8>,
'opt_algo': 'adam',
'prefetch_size': 6,
'pretrain': None,
'result_dir': 'results',
'sampling_function': 'kp_detection',
'snapshot': 5000,
'snapshot_name': 'CenterNet-104',
'stepsize': 4500,
'test_split': 'testdev',
'train_split': 'trainval',
'val_iter': 200,
'val_split': 'minival',
'weight_decay': False,
'weight_decay_rate': 1e-05,
'weight_decay_type': 'l2'}
db config...
{'ae_threshold': 0.5,
'border': 128,
'categories': 1,
'data_aug': True,
'gaussian_bump': True,
'gaussian_iou': 0.7,
'gaussian_radius': -1,
'input_size': [511, 511],
'kp_categories': 1,
'lighting': True,
'max_per_image': 100,
'merge_bbox': False,
'nms_algorithm': 'exp_soft_nms',
'nms_kernel': 3,
'nms_threshold': 0.5,
'output_sizes': [[128, 128]],
'rand_color': True,
'rand_crop': True,
'rand_pushes': False,
'rand_samples': False,
'rand_scale_max': 1.4,
'rand_scale_min': 0.6,
'rand_scale_step': 0.1,
'rand_scales': array([0.6, 0.7, 0.8, 0.9, 1. , 1.1, 1.2, 1.3]),
'special_crop': False,
'test_scales': [1],
'top_k': 70,
'weight_exp': 8}
len of db: 1451
start prefetching data...
shuffling indices...
building model...
module_file: models.CenterNet-104
start prefetching data...
shuffling indices...
total parameters: 210062960
setting learning rate to: 0.00025
training start...
0%| | 0/100000 [00:00<?, ?it/s]THCudaCheck FAIL file=/opt/conda/conda-bld/pytorch_1535491974311/work/aten/src/THC/THCGeneral.cpp line=663 error=11 : invalid argument
/home/imdl/anaconda3/envs/centernet/lib/python3.6/site-packages/torch/nn/modules/upsampling.py:122: UserWarning: nn.Upsampling is deprecated. Use nn.functional.interpolate instead.
warnings.warn("nn.Upsampling is deprecated. Use nn.functional.interpolate instead.")
Traceback (most recent call last):
File "/home/imdl/workspace/CenterNet-master_train_custom_dataset/train.py", line 204, in
train(training_dbs, validation_db, args.start_iter)
File "/home/imdl/workspace/CenterNet-master_train_custom_dataset/train.py", line 138, in train
training_loss, focal_loss, pull_loss, push_loss, regr_loss = nnet.train(**training)
File "/home/imdl/workspace/CenterNet-master_train_custom_dataset/nnet/py_factory.py", line 82, in train
loss_kp = self.network(xs, ys)
File "/home/imdl/anaconda3/envs/centernet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/home/imdl/workspace/CenterNet-master_train_custom_dataset/models/py_utils/data_parallel.py", line 68, in forward
return self.module(*inputs[0], **kwargs[0])
File "/home/imdl/anaconda3/envs/centernet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/home/imdl/workspace/CenterNet-master_train_custom_dataset/nnet/py_factory.py", line 21, in forward
loss_kp = self.loss(preds, ys, **kwargs)
File "/home/imdl/anaconda3/envs/centernet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/home/imdl/workspace/CenterNet-master_train_custom_dataset/models/py_utils/kp.py", line 330, in forward
focal_loss += self.focal_loss(tl_heats, gt_tl_heat)
File "/home/imdl/workspace/CenterNet-master_train_custom_dataset/models/py_utils/kp_utils.py", line 180, in _neg_loss
pos_pred = pred[pos_inds]
RuntimeError: The shape of the mask [1, 1, 128, 128] at index 1 does not match the shape of the indexed tensor [1, 80, 128, 128] at index 1

Process finished with exit code 1

[li10141110]

I replaced the content of my .data/coco/annotations/instances_{minival2014,trainval2014}.json to my coco style file and replaced the content of my .data/coco/images/{minival2014,trainval2014} to my own images.
following is my CenterNet-104.json file.
{
"system": {
"dataset": "MSCOCO",
"batch_size": 1,
"sampling_function": "kp_detection",

    "train_split": "trainval",
    "val_split": "minival",

    "learning_rate": 0.00025,
    "decay_rate": 10,

    "val_iter": 200,

    "opt_algo": "adam",
    "prefetch_size": 6,

    "max_iter": 100000,
    "stepsize": 4500,
    "snapshot": 5000,

    "chunk_sizes": [1],

    "data_dir": "../data"
},

"db": {
    "rand_scale_min": 0.6,
    "rand_scale_max": 1.4,
    "rand_scale_step": 0.1,
    "rand_scales": null,

    "rand_crop": true,
    "rand_color": true,

    "border": 128,
    "gaussian_bump": true,

    "input_size": [511, 511],
    "output_sizes": [[128, 128]],

    "test_scales": [1],

    "top_k": 70,
    "categories": 1,
    "kp_categories": 1,
    "ae_threshold": 0.5,
    "nms_threshold": 0.5,

    "max_per_image": 100
}

}

[Duankaiwen]

@li10141110 How many categories do you have？

[Duankaiwen]

@li10141110 Add these codes above the
dist = tag_mean.unsqueeze(1) - tag_mean.unsqueeze(2)
in models/py_utils/kp_utils.py:

if len(tag_mean.size()) < 2:

      tag_mean = tag_mean.unsqueeze(0)`  

[li10141110]

1 category

[li10141110]

Thank you Kaiwen, I make it

[nassarofficial]

Did it work with you? I had the same problem but the solution was to edit the categories number in the config file. I am also training with 1 class but I am having problems in the evaluation.

I have only 1 class, and it's giving me a "key error 0"

        category       = self._coco_to_class_map[cat_id]

in the db/coco.py file

[deprecatedwarrior]

In models/CenterNet-104.py, model class definition, the out_dim is hard-coded to 80. For custom datasets with a different out_dim, these need to be changed along with the above-mentioned changes. Just making a note for other users.