Testing this on my own dataset about centernet HOT 20 OPEN

@chariot1498 Hi, you can modify the "kp_dectection" function in test/coco.py.

from centernet.

@Duankaiwen I have trained one category object detection centernet based on your project but I found that in test.py my model can only get low scores(about 0.4-0.6). It confused me cause I can obtain 0.46AP and 0.828 [email protected]. Followed is my test result and screenshot of code

test @ 6000
/home/imdl/anaconda3/envs/centernet/bin/python /home/imdl/workspace/CenterNet-master_train_custom_dataset/test.py CenterNet-52 --testiter 6000 --split validation
cfg_file: config/CenterNet-52.json
loading all datasets...
split: minival
loading from cache file: cache/coco_minival2014.pkl
loading annotations into memory...
Done (t=0.00s)
creating index...
index created!
system config...
{'batch_size': 16,
'cache_dir': 'cache',
'chunk_sizes': [16],
'config_dir': 'config',
'data_dir': '../data',
'data_rng': <mtrand.RandomState object at 0x7f7560b8c708>,
'dataset': 'MSCOCO',
'decay_rate': 10,
'display': 5,
'learning_rate': 0.0005,
'max_iter': 40000,
'nnet_rng': <mtrand.RandomState object at 0x7f7560b8c750>,
'opt_algo': 'adam',
'prefetch_size': 6,
'pretrain': None,
'result_dir': 'results',
'sampling_function': 'kp_detection',
'snapshot': 1000,
'snapshot_name': 'CenterNet-52',
'stepsize': 10000,
'test_split': 'testdev',
'train_split': 'trainval',
'val_iter': 500,
'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': 50,
'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}
loading parameters at iteration: 6000
building neural network...
module_file: models.CenterNet-52
total parameters: 104783243
loading parameters...
loading model from cache/nnet/CenterNet-52/CenterNet-52_6000.pkl
locating kps: 0%| | 0/348 [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.")
locating kps: 100%|█| 348/348 [01:06<00:00, 5.36it/s]
Loading and preparing results...
DONE (t=0.04s)
creating index...
index created!
Running per image evaluation...
Evaluate annotation type bbox
DONE (t=1.05s).
Accumulating evaluation results...
DONE (t=0.09s).
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.461
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.828
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.487
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.060
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.484
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = -1.000
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.456
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.591
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.629
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.500
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.629
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = -1.000
Running per image evaluation...
Evaluate annotation type bbox
DONE (t=0.99s).
Accumulating evaluation results...
DONE (t=0.09s).
False discovery (FD) @[ IoU=0.05:0.50 | area= all | maxDets=100 ] = 0.161
False discovery (FD) @[ IoU=0.05 | area= all | maxDets=100 ] = 0.153
False discovery (FD) @[ IoU=0.25 | area= all | maxDets=100 ] = 0.159
False discovery (FD) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.172
False discovery (FD) @[ IoU=0.05:0.50 | area= small | maxDets=100 ] = 0.790
False discovery (FD) @[ IoU=0.05:0.50 | area=medium | maxDets=100 ] = 0.120
False discovery (FD) @[ IoU=0.05:0.50 | area= large | maxDets=100 ] = -1.000

Process finished with exit code 0

from centernet.

@li10141110 你好，我训练ok，test出问题：
(CenterNet) [root@add CenterNet-master]# python test.py CenterNet-104 --testiter 12000 --split validation
cfg_file: config/CenterNet-104.json
loading all datasets...
split: minival
minival
{1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9, 10: 10, 11: 11, 12: 13, 13: 14, 14: 15, 15: 16, 16: 17, 17: 18, 18: 19, 19: 20, 20: 21, 21: 22, 22: 23, 23: 24, 24: 25, 25: 27, 26: 28, 27: 31, 28: 32, 29: 33, 30: 34, 31: 35, 32: 36, 33: 37, 34: 38, 35: 39, 36: 40, 37: 41, 38: 42, 39: 43, 40: 44, 41: 46, 42: 47, 43: 48, 44: 49, 45: 50, 46: 51, 47: 52, 48: 53, 49: 54, 50: 55, 51: 56, 52: 57, 53: 58, 54: 59, 55: 60, 56: 61, 57: 62, 58: 63, 59: 64, 60: 65, 61: 67, 62: 70, 63: 72, 64: 73, 65: 74, 66: 75, 67: 76, 68: 77, 69: 78, 70: 79, 71: 80, 72: 81, 73: 82, 74: 84, 75: 85, 76: 86, 77: 87, 78: 88, 79: 89, 80: 90}
loading from cache file: cache/coco_val.pkl
loading annotations into memory...
Done (t=0.00s)
creating index...
index created!
<db.coco.MSCOCO object at 0x7f04291065c0>
system config...
{'batch_size': 2,
'cache_dir': 'cache',
'chunk_sizes': [2],
'config_dir': 'config',
'data_dir': '/data2/hejin/CenterNet-master/data',
'data_rng': <mtrand.RandomState object at 0x7f0460ec1ab0>,
'dataset': 'MSCOCO',
'decay_rate': 10,
'display': 5,
'learning_rate': 0.00025,
'max_iter': 12000,
'nnet_rng': <mtrand.RandomState object at 0x7f0460ec1af8>,
'opt_algo': 'adam',
'prefetch_size': 6,
'pretrain': None,
'result_dir': 'results',
'sampling_function': 'kp_detection',
'snapshot': 500,
'snapshot_name': 'CenterNet-104',
'stepsize': 5000,
'test_split': 'testdev',
'train_split': 'trainval',
'val_iter': 500,
'val_split': 'minival',
'weight_decay': False,
'weight_decay_rate': 1e-05,
'weight_decay_type': 'l2'}
db config...
{'ae_threshold': 0.5,
'border': 128,
'categories': 3,
'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': False,
'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}
loading parameters at iteration: 12000
building neural network...
module_file: models.CenterNet-104
total parameters: 209944226
loading parameters...
loading model from cache/nnet/CenterNet-104/CenterNet-104_12000.pkl
test.coco
<function testing at 0x7f0428af2158>
locating kps: 0%| | 0/10 [00:00<?, ?it/s]Traceback (most recent call last):
File "test.py", line 102, in
test(testing_db, args.split, args.testiter, args.debug, args.suffix)
File "test.py", line 62, in test
testing(db, nnet, result_dir, debug=debug)
File "/data2/hejin/CenterNet-master/test/coco.py", line 321, in testing
return globals()[system_configs.sampling_function](db, nnet, result_dir, debug=debug)
File "/data2/hejin/CenterNet-master/test/coco.py", line 129, in kp_detection
dets, center = decode_func(nnet, images, K, ae_threshold=ae_threshold, kernel=nms_kernel)
File "/data2/hejin/CenterNet-master/test/coco.py", line 54, in kp_decode
detections, center = nnet.test([images], ae_threshold=ae_threshold, K=K, kernel=kernel)
File "/data2/hejin/CenterNet-master/nnet/py_factory.py", line 114, in test
return self.model(*xs, **kwargs)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/data2/hejin/CenterNet-master/nnet/py_factory.py", line 32, in forward
return self.module(*xs, **kwargs)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/data2/hejin/CenterNet-master/models/py_utils/kp.py", line 290, in forward
return self._test(*xs, **kwargs)
File "/data2/hejin/CenterNet-master/models/py_utils/kp.py", line 243, in _test
inter = self.pre(image)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/container.py", line 91, in forward
input = module(input)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/data2/hejin/CenterNet-master/models/py_utils/utils.py", line 15, in forward
bn = self.bn(conv)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/batchnorm.py", line 66, in forward
exponential_average_factor, self.eps)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/functional.py", line 1254, in batch_norm
training, momentum, eps, torch.backends.cudnn.enabled
RuntimeError: CUDA error: out of memory

from centernet.

your GPU memory is limited.

from centernet.

set batch size to 1 and chunk_size [1]

from centernet.

I want to train my own dataset and I format the data to the coco format, change the class_id in the ./db/coco.py, where else should I modify? Thanks

from centernet.

@li10141110 what's the chunk_sizes means? how can i set gpu id in this project?

from centernet.

@li10141110 Hi, the scores only affect the order of boxes.

from centernet.

@NightNightNight chunk_sizes denotes the batch size for each GPU

from centernet.

@Duankaiwen thanks for you reply, i know the means of chunk_sizes now.

from centernet.

@li10141110 你好，我训练ok，test出问题：
(CenterNet) [root@add CenterNet-master]# python test.py CenterNet-104 --testiter 12000 --split validation
cfg_file: config/CenterNet-104.json
loading all datasets...
split: minival
minival
{1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9, 10: 10, 11: 11, 12: 13, 13: 14, 14: 15, 15: 16, 16: 17, 17: 18, 18: 19, 19: 20, 20: 21, 21: 22, 22: 23, 23: 24, 24: 25, 25: 27, 26: 28, 27: 31, 28: 32, 29: 33, 30: 34, 31: 35, 32: 36, 33: 37, 34: 38, 35: 39, 36: 40, 37: 41, 38: 42, 39: 43, 40: 44, 41: 46, 42: 47, 43: 48, 44: 49, 45: 50, 46: 51, 47: 52, 48: 53, 49: 54, 50: 55, 51: 56, 52: 57, 53: 58, 54: 59, 55: 60, 56: 61, 57: 62, 58: 63, 59: 64, 60: 65, 61: 67, 62: 70, 63: 72, 64: 73, 65: 74, 66: 75, 67: 76, 68: 77, 69: 78, 70: 79, 71: 80, 72: 81, 73: 82, 74: 84, 75: 85, 76: 86, 77: 87, 78: 88, 79: 89, 80: 90}
loading from cache file: cache/coco_val.pkl
loading annotations into memory...
Done (t=0.00s)
creating index...
index created!
<db.coco.MSCOCO object at 0x7f04291065c0>
system config...
{'batch_size': 2,
'cache_dir': 'cache',
'chunk_sizes': [2],
'config_dir': 'config',
'data_dir': '/data2/hejin/CenterNet-master/data',
'data_rng': <mtrand.RandomState object at 0x7f0460ec1ab0>,
'dataset': 'MSCOCO',
'decay_rate': 10,
'display': 5,
'learning_rate': 0.00025,
'max_iter': 12000,
'nnet_rng': <mtrand.RandomState object at 0x7f0460ec1af8>,
'opt_algo': 'adam',
'prefetch_size': 6,
'pretrain': None,
'result_dir': 'results',
'sampling_function': 'kp_detection',
'snapshot': 500,
'snapshot_name': 'CenterNet-104',
'stepsize': 5000,
'test_split': 'testdev',
'train_split': 'trainval',
'val_iter': 500,
'val_split': 'minival',
'weight_decay': False,
'weight_decay_rate': 1e-05,
'weight_decay_type': 'l2'}
db config...
{'ae_threshold': 0.5,
'border': 128,
'categories': 3,
'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': False,
'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}
loading parameters at iteration: 12000
building neural network...
module_file: models.CenterNet-104
total parameters: 209944226
loading parameters...
loading model from cache/nnet/CenterNet-104/CenterNet-104_12000.pkl
test.coco
<function testing at 0x7f0428af2158>
locating kps: 0%| | 0/10 [00:00<?, ?it/s]Traceback (most recent call last):
File "test.py", line 102, in
test(testing_db, args.split, args.testiter, args.debug, args.suffix)
File "test.py", line 62, in test
testing(db, nnet, result_dir, debug=debug)
File "/data2/hejin/CenterNet-master/test/coco.py", line 321, in testing
return globals()[system_configs.sampling_function](db, nnet, result_dir, debug=debug)
File "/data2/hejin/CenterNet-master/test/coco.py", line 129, in kp_detection
dets, center = decode_func(nnet, images, K, ae_threshold=ae_threshold, kernel=nms_kernel)
File "/data2/hejin/CenterNet-master/test/coco.py", line 54, in kp_decode
detections, center = nnet.test([images], ae_threshold=ae_threshold, K=K, kernel=kernel)
File "/data2/hejin/CenterNet-master/nnet/py_factory.py", line 114, in test
return self.model(*xs, **kwargs)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/data2/hejin/CenterNet-master/nnet/py_factory.py", line 32, in forward
return self.module(*xs, **kwargs)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/data2/hejin/CenterNet-master/models/py_utils/kp.py", line 290, in forward
return self._test(*xs, **kwargs)
File "/data2/hejin/CenterNet-master/models/py_utils/kp.py", line 243, in _test
inter = self.pre(image)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/container.py", line 91, in forward
input = module(input)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/data2/hejin/CenterNet-master/models/py_utils/utils.py", line 15, in forward
bn = self.bn(conv)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/modules/batchnorm.py", line 66, in forward
exponential_average_factor, self.eps)
File "/data2/hejin/anaconda2/envs/CenterNet/lib/python3.6/site-packages/torch/nn/functional.py", line 1254, in batch_norm
training, momentum, eps, torch.backends.cudnn.enabled
RuntimeError: CUDA error: out of memory

你好，我也遇到了这个问题，训练自己的数据集的时候，训练可以，但是测试就显存不足，请问怎么解决？

from centernet.

from centernet.

你好，请问一下，当我缩小图片的时候，运行test.py时会出现ValueError: need at least one array to concatenate，这个原因是np.concatenate(center_points, axis=1)中center_points是空集，可能原因是缩小图片会出现没有GT的情况，那我应该怎么解决这个问题?谢谢答疑

from centernet.

@yang-chenyu104
你修改scales的值了么？

from centernet.

from centernet.

谢谢解答，我把scaless修改了，有个问题不懂，为啥要加一个scales==1的判断，把那行注释了#if scale == 1:目前正在生成results.json，谢谢大佬的代码,对我帮助很大!!!,想用中心点检测去做表格方面的研究，希望在自己研究方面加深!!!
…
------------------ 原始邮件 ------------------ 发件人: "Duankaiwen/CenterNet" @.>; 发送时间: 2021年5月28日(星期五) 下午3:42 @.>; 抄送: "な法--言说 @.@.>; 主题: Re: [Duankaiwen/CenterNet] Testing this on my own dataset (#21) @yang-chenyu104 你修改scales的值了么？ — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

@yang-chenyu104
实验时发现只需要scale=1时检测到的center就够了，因此只取了scale=1时的center。希望对你有帮助。

from centernet.

from centernet.

from centernet.

在代码里面我怎么去找寻效果最好的一次的损失和正确率
…
------------------ 原始邮件 ------------------ 发件人: "Duankaiwen/CenterNet" @.>; 发送时间: 2021年5月28日(星期五) 下午4:00 @.>; 抄送: "な法--言说 @.@.>; 主题: Re: [Duankaiwen/CenterNet] Testing this on my own dataset (#21) 谢谢解答，我把scaless修改了，有个问题不懂，为啥要加一个scales==1的判断，把那行注释了#if scale == 1:目前正在生成results.json，谢谢大佬的代码,对我帮助很大!!!,想用中心点检测去做表格方面的研究，希望在自己研究方面加深!!! … ------------------ 原始邮件 ------------------ 发件人: "Duankaiwen/CenterNet" @.>; 发送时间: 2021年5月28日(星期五) 下午3:42 @.>; 抄送: "な法--言说 @.@.>; 主题: Re: [Duankaiwen/CenterNet] Testing this on my own dataset (#21) @yang-chenyu104 你修改scales的值了么？ — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe. @yang-chenyu104 实验时发现只需要scale=1时检测到的center就够了，因此只取了scale=1时的center。希望对你有帮助。 — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

@yang-chenyu104

每次iter只有loss，等训练结束之后才统一测检测精度，所以没有办法去找效果最好的一次的正确率，你倒是可以找到1w，2w，..., 48w iter，这48个中最好的一次精度，因为每隔10000个iter会保存一次模型，比如你训练了480000 iters，那么你就有48个模型，加载每个模型测一下，找出最好的精度.

在trian.py line 138和line 154有每个iter之后的loss，你可以统计一下，找到最好的一次loss

from centernet.

from centernet.