arashwan / matrixnet Goto Github PK

why the the base layer (top left) is set as [24,48,24,48]
but in corresonding layer range the the base layer (top left) is [0,48,0,48]

I ran train.py with my own images with 5 categories. It was done.
Now, I am trying to run test.py.
$ python test.py MatrixNetCornersResnet50 --testiter 340000 --split validation

Unfortunately, I am getting error messages. I will really appreciate if you can show me how to fix the error.

This is the log.

cfg_file: ./config/MatrixNetCornersResnet50.json
loading all datasets...
split: test
loading from cache file: ./MatrixNetCornersResnet50_48LayerRange_512isize/coco_test.pkl
loading annotations into memory...
Done (t=0.01s)
creating index...
index created!
system config...
{'batch_size': 23,
'cache_dir': './MatrixNetCornersResnet50_48LayerRange_512isize',
'chunk_sizes': [5, 6, 6, 6],
'config_dir': './config',
'data_dir': './data/',
'data_rng': RandomState(MT19937) at 0x2ADE183A2048,
'dataset': 'MSCOCO',
'decay_rate': 10,
'display': 5,
'learning_rate': 5e-05,
'max_iter': 350000,
'model_name': 'MatrixNetCorners',
'nnet_rng': RandomState(MT19937) at 0x2ADE183A2150,
'opt_algo': 'adam',
'prefetch_size': 6,
'pretrain': None,
'result_dir': './results',
'snapshot': 10000,
'stepsize': 300000,
'test_split': 'testdev2017',
'train_split': 'train',
'val_iter': 100,
'val_split': 'test',
'weight_decay': False,
'weight_decay_rate': 1e-05,
'weight_decay_type': 'l2'}
db config...
{'backbone': 'resnet50',
'base_layer_range': [24, 48, 24, 48],
'border': 128,
'categories': 5,
'cutout': True,
'data_aug': True,
'gaussian_bump': True,
'gaussian_iou': 0.3,
'gaussian_radius': -1,
'height_thresholds': None,
'input_size': [512, 512],
'layers_range': [[[0, 48, 0, 48], [48, 96, 0, 48], [96, 192, 0, 48], -1, -1],
[[0, 48, 48, 96],
[48, 96, 48, 96],
[96, 192, 48, 96],
[192, 384, 0, 96],
-1],
[[0, 48, 96, 192],
[48, 96, 96, 192],
[96, 192, 96, 192],
[192, 384, 96, 192],
[384, 2000, 96, 192]],
[-1,
[0, 96, 192, 384],
[96, 192, 192, 384],
[192, 384, 192, 384],
[384, 2000, 192, 384]],
[-1,
-1,
[0, 192, 384, 2000],
[192, 384, 384, 2000],
[384, 2000, 384, 2000]]],
'lighting': True,
'matching_threshold': 0.3,
'max_per_image': 100,
'merge_bbox': True,
'nms_algorithm': 'exp_soft_nms',
'nms_kernel': 3,
'nms_threshold': 0.5,
'output_kernel_size': 3,
'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_flip_images': True,
'test_image_max_dim': 900,
'test_scales': [1],
'top_k': 50,
'train_image_max_dim': 800,
'weight_exp': 6,
'width_thresholds': None}
loading parameters at iteration: 340000
building neural network...
module_file: models.MatrixNetCorners
total parameters: 48347962
loading parameters...
loading model from ./MatrixNetCornersResnet50_48LayerRange_512isize/nnet/MatrixNetCorners/MatrixNetCorners_340000.pkl

However, I am getting the following error message.

locating kps: 0%| | 0/575 [00:00<?, ?it/s]
Traceback (most recent call last):
File "test.py", line 98, in
test(testing_db, args.split, args.testiter, args.debug, args.suffix)
File "test.py", line 61, in test
testing(db, nnet, result_dir, debug=debug)
File "/data/hsyi/Pray/matrixnet/test/coco.py", line 433, in testing
return globals()["test_"+system_configs.model_name](db, nnet, result_dir, debug=debug)
File "/data/hsyi/Pray/matrixnet/test/coco.py", line 164, in test_MatrixNetCorners
soft_nms_merge(top_bboxes[image_id][j + 1], Nt=nms_threshold, method=nms_algorithm, weight_exp=weight_exp)
File "nms.pyx", line 280, in nms.soft_nms_merge
ZeroDivisionError: float division

Thank you in advance.

@rra94 results and src directories are still there, I thought we got rid of them. Did you forget to push last commit?

could you release demo code to test on arbitrary images?

Hello, guys, I am confused about why regressing the centers directly in corner-xnet can replace the embeddings in cornerNet to do the corner matching ?

could you give a baidu cloud link for the pretrained model,thanks

see title

i try to run my dataset to train, anno & img dir put in data/coco/annotations & images like u in readme

it print "training start..." but stuck at 0%

and i find that some error in samples/coco.py in cutout
at center_random = [random.randint(0, image.shape[0]), random.randint(0, image.shape[1])]
error message AttributeError: 'NoneType' object has no attribute 'shape'

i want to know this error is my dataset maybe wrong or somewhere i need to change

thanks~