RocketGit

hahattpro / crowd_counting_framework (public) (License: Unspecified) (since 2020-12-23) (hash sha1)

backupfrom https://github.com/ttpro1995/crowd_counting_framework

Clone URLs: https://rocketgit.com/user/hahattpro/crowd_counting_framework ssh://rocketgit@ssh.rocketgit.com/user/hahattpro/crowd_counting_framework git://git.rocketgit.com/user/hahattpro/crowd_counting_framework

List of commits:

Subject	Hash	Author	Date (UTC)
setting up code	6e83edd7283bfeec6ec1a6e5057719e8ae95e8ea	Thai Thien	2020-04-02 10:25:18
fix typo	f7565a25dc7058d1f344cc0d6e610fa2bf8f19e9	Thai Thien	2020-04-01 10:58:38
t4 aug	36b5bbc17111dca87cfa082f5a512b84576c8fca	Thai Thien	2020-04-01 10:55:12
reproduce ccnn without batchnorm	e4678b328ba9d00b9d94617a1cd0f34732f7b2ac	Thai Thien	2020-03-30 16:48:35
change gpu number	f4c616536d13d4caf383166caae0d8298cb73738	Thai Thien	2020-03-28 04:17:37
train custom ccnn_v2	8b6fca9cd3cb15111befd7332bc7e764d5494a66	Thai Thien	2020-03-28 03:51:07
v5	24e4a8a9c8948c21b0d6beadec0418adceec8577	Thai Thien	2020-03-20 17:56:32
hope it fix cuda bug	7b8cbb70550df279035ef56a22450d17fed6b29a	Thai Thien	2020-03-20 17:55:24
ccnnv4 sha	a4acad0caf5a18fc9ea037a5e3d6db9d6b6106c7	Thai Thien	2020-03-20 17:38:30
change gpu	fa2c1c54858d8a57585c3801dc168c464ba689dd	Thai Thien	2020-03-18 17:31:38
fix typo again	d71ce17c4efbcc2cd31917f0e27e90bd964ac964	Thai Thien	2020-03-18 17:30:34
typo again	c16871fb2cc035e168d62df9c5d4401bc465df95	Thai Thien	2020-03-18 17:27:45
fix minor typo	01d907dda5c1e4c08b5ca5607aa77365cfd70a2e	Thai Thien	2020-03-18 17:25:32
forgot to change the model	cb8bde150d7e764f67f33c980759675a2f5cce8e	Thai Thien	2020-03-18 17:23:51
log summary for adcrowdnet simple lrscheduler	cefb188f185dbba3dd95f5034f17bd43dd194d44	Thai Thien	2020-03-18 17:18:36
prepare exp	4658d0a852d952f1e1f600d9c494b8d511b9d52c	Thai Thien	2020-03-18 17:17:05
CompactCNNV6 with a bunch of batchnorm	a9f21eac8ab6a328fac7a23bb5eddbbaa4496d04	Thai Thien	2020-03-18 17:09:26
implement attn_can_adcrowdnet_v5	c905bd76d3a5d3b933d1e6eab6e50da187cee3a2	Thai Thien	2020-03-18 16:56:02
forget add --momentum	58131da12819203ec13ea62e7ad6862cf6db301d	thient	2020-03-18 03:41:30
shell script	ebb46f6853a666ecdfbaf42bb927cad0ec9e548e	thient	2020-03-18 03:38:56

Commit 6e83edd7283bfeec6ec1a6e5057719e8ae95e8ea - setting up code
Author: Thai Thien
Author date (UTC): 2020-04-02 10:25
Committer name: Thai Thien
Committer date (UTC): 2020-04-02 10:25
Parent(s): f7565a25dc7058d1f344cc0d6e610fa2bf8f19e9
Signer:
Signing key:
Signing status: N
Tree: f7ac71a6644180c8a54c15ce2abb5c8769ce5cf0

File	Lines added	Lines deleted
args_util.py	21	0
experiment_meow_main.py	13	5
models/meow_experiment/__init__.py	0	0
models/meow_experiment/kitten_meow_1.py	119	0

File args_util.py changed (mode: 100644) (index bbe5616..5f8c6f4)
...	...	def my_args_parse():
100	100	return arg	return arg
101	101
102	102
	103		def meow_parse():
	104		parser = argparse.ArgumentParser(description='CrowdCounting Context Aware Network')
	105		parser.add_argument("--task_id", action="store", default="dev")
	106		parser.add_argument("--model", action="store", default="dev")
	107		parser.add_argument('--note', action="store", default="write anything")
	108
	109		parser.add_argument('--input', action="store", type=str, default=HardCodeVariable().SHANGHAITECH_PATH_PART_A)
	110		parser.add_argument('--datasetname', action="store", default="shanghaitech_keepfull")
	111
	112		# args with default value
	113		parser.add_argument('--load_model', action="store", default="", type=str)
	114		parser.add_argument('--lr', action="store", default=1e-8, type=float)
	115		parser.add_argument('--momentum', action="store", default=0.9, type=float)
	116		parser.add_argument('--decay', action="store", default=5*1e-3, type=float)
	117		parser.add_argument('--epochs', action="store", default=1, type=int)
	118		parser.add_argument('--batch_size', action="store", default=1, type=int,
	119		help="only set batch_size > 0 for dataset with image size equal")
	120		parser.add_argument('--test', action="store_true", default=False)
	121		arg = parser.parse_args()
	122		return arg
	123
103	124	def sanity_check_dataloader_parse():	def sanity_check_dataloader_parse():
104	125	parser = argparse.ArgumentParser(description='Dataloader')	parser = argparse.ArgumentParser(description='Dataloader')
105	126	parser.add_argument('--input', action="store", type=str, default=HardCodeVariable().SHANGHAITECH_PATH_PART_A)	parser.add_argument('--input', action="store", type=str, default=HardCodeVariable().SHANGHAITECH_PATH_PART_A)

File experiment_meow_main.py copied from file train_custom_compact_cnn.py (similarity 92%) (mode: 100644) (index 07a23b0..7b76dc2)
1	1	from comet_ml import Experiment	from comet_ml import Experiment
2	2
3		from args_util import my_args_parse
	3		from args_util import meow_parse
4	4	from data_flow import get_train_val_list, get_dataloader, create_training_image_list, create_image_list	from data_flow import get_train_val_list, get_dataloader, create_training_image_list, create_image_list
5	5	from ignite.engine import Events, create_supervised_trainer, create_supervised_evaluator	from ignite.engine import Events, create_supervised_trainer, create_supervised_evaluator
6	6	from ignite.metrics import Loss, MeanAbsoluteError, MeanSquaredError	from ignite.metrics import Loss, MeanAbsoluteError, MeanSquaredError

...	...	from visualize_util import get_readable_time
11	11
12	12	import torch	import torch
13	13	from torch import nn	from torch import nn
14		from models import CustomCNNv2
	14		from models.meow_experiment.kitten_meow_1 import M1, M2
15	15	import os	import os
16	16	from model_util import get_lr	from model_util import get_lr
17	17
18	18	COMET_ML_API = "S3mM1eMq6NumMxk2QJAXASkUM"	COMET_ML_API = "S3mM1eMq6NumMxk2QJAXASkUM"
19		PROJECT_NAME = "crowd-counting-framework"
	19		PROJECT_NAME = "meow-one-experiment-insita"
20	20
21	21	def very_simple_param_count(model):	def very_simple_param_count(model):
22	22	result = sum([p.numel() for p in model.parameters()])	result = sum([p.numel() for p in model.parameters()])

...	...	if __name__ == "__main__":
26	26	experiment = Experiment(project_name=PROJECT_NAME, api_key=COMET_ML_API)	experiment = Experiment(project_name=PROJECT_NAME, api_key=COMET_ML_API)
27	27	device = torch.device("cuda" if torch.cuda.is_available() else "cpu")	device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
28	28	print(device)	print(device)
29		args = my_args_parse()
	29		args = meow_parse()
30	30	print(args)	print(args)
31	31
32	32	experiment.set_name(args.task_id)	experiment.set_name(args.task_id)

...	...	if __name__ == "__main__":
55	55	print("len train_loader ", len(train_loader))	print("len train_loader ", len(train_loader))
56	56
57	57	# model	# model
58		model = CustomCNNv2()
	58		model_name = args.model
	59		experiment.log_other("model", model_name)
	60		if model_name == "M1":
	61		model = M1()
	62		elif model_name == "M2":
	63		model = M2()
	64		else:
	65		print("error: you didn't pick a model")
	66		exit(-1)
59	67	n_param = very_simple_param_count(model)	n_param = very_simple_param_count(model)
60	68	experiment.log_other("n_param", n_param)	experiment.log_other("n_param", n_param)
61	69	if hasattr(model, 'model_note'):	if hasattr(model, 'model_note'):

File models/meow_experiment/__init__.py copied from file playground/__init__.py (similarity 100%)

File models/meow_experiment/kitten_meow_1.py added (mode: 100644) (index 0000000..055ce5f)
	1	import torch.nn as nn
	2	import torch
	3	from torchvision import models
	4	from models.deform_conv_v2 import DeformConv2d, TorchVisionBasicDeformConv2d
	5	import collections
	6	import torch.nn.functional as F
	7
	8
	9	class M1(nn.Module):
	10	"""
	11	A REAL-TIME DEEP NETWORK FOR CROWD COUNTING
	12	https://arxiv.org/pdf/2002.06515.pdf
	13	the improve version
	14
	15	we change 5x5 7x7 9x9 with 3x3
	16	Keep the tail
	17	"""
	18	def __init__(self, load_weights=False):
	19	super(M1, self).__init__()
	20	self.model_note = "We replace 5x5 7x7 9x9 with 3x3, no batchnorm yet, keep tail, no dilated"
	21	# self.red_cnn = nn.Conv2d(3, 10, 9, padding=4)
	22	# self.green_cnn = nn.Conv2d(3, 14, 7, padding=3)
	23	# self.blue_cnn = nn.Conv2d(3, 16, 5, padding=2)
	24
	25	# ideal from crowd counting using DMCNN
	26	self.front_cnn_1 = nn.Conv2d(3, 20, 3, padding=1)
	27	self.front_cnn_2 = nn.Conv2d(20, 16, 3, padding=1)
	28	self.front_cnn_3 = nn.Conv2d(16, 14, 3, padding=1)
	29	self.front_cnn_4 = nn.Conv2d(14, 10, 3, padding=1)
	30
	31	self.c0 = nn.Conv2d(40, 40, 3, padding=1)
	32	self.max_pooling = nn.MaxPool2d(kernel_size=2, stride=2)
	33
	34	self.c1 = nn.Conv2d(40, 60, 3, padding=1)
	35
	36	# ideal from CSRNet
	37	self.c2 = nn.Conv2d(60, 40, 3, padding=1)
	38	self.c3 = nn.Conv2d(40, 20, 3, padding=1)
	39	self.c4 = nn.Conv2d(20, 10, 3, padding=1)
	40	self.output = nn.Conv2d(10, 1, 1)
	41
	42	def forward(self,x):
	43	#x_red = self.max_pooling(F.relu(self.red_cnn(x), inplace=True))
	44	#x_green = self.max_pooling(F.relu(self.green_cnn(x), inplace=True))
	45	#x_blue = self.max_pooling(F.relu(self.blue_cnn(x), inplace=True))
	46
	47	x_red = F.relu(self.front_cnn_1(x), inplace=True)
	48	x_red = F.relu(self.front_cnn_2(x_red), inplace=True)
	49	x_red = F.relu(self.front_cnn_3(x_red), inplace=True)
	50	x_red = F.relu(self.front_cnn_4(x_red), inplace=True)
	51	x_red = self.max_pooling(x_red)
	52
	53	x_green = F.relu(self.front_cnn_1(x), inplace=True)
	54	x_green = F.relu(self.front_cnn_2(x_green), inplace=True)
	55	x_green = F.relu(self.front_cnn_3(x_green), inplace=True)
	56	x_green = self.max_pooling(x_green)
	57
	58	x_blue = F.relu(self.front_cnn_1(x), inplace=True)
	59	x_blue = F.relu(self.front_cnn_2(x_blue), inplace=True)
	60	x_blue = self.max_pooling(x_blue)
	61
	62	x = torch.cat((x_red, x_green, x_blue), 1)
	63	x = F.relu(self.c0(x), inplace=True)
	64
	65	x = F.relu(self.c1(x), inplace=True)
	66
	67	x = F.relu(self.c2(x), inplace=True)
	68	x = self.max_pooling(x)
	69
	70	x = F.relu(self.c3(x), inplace=True)
	71	x = self.max_pooling(x)
	72
	73	x = F.relu(self.c4(x), inplace=True)
	74
	75	x = self.output(x)
	76	return x
	77
	78
	79	class M2(nn.Module):
	80	"""
	81	A REAL-TIME DEEP NETWORK FOR CROWD COUNTING
	82	https://arxiv.org/pdf/2002.06515.pdf
	83	"""
	84	def __init__(self, load_weights=False):
	85	super(M2, self).__init__()
	86	self.model_note = "No batchnorm, keep head, but dilated tail"
	87	self.red_cnn = nn.Conv2d(3, 10, 9, padding=4)
	88	self.green_cnn = nn.Conv2d(3, 14, 7, padding=3)
	89	self.blue_cnn = nn.Conv2d(3, 16, 5, padding=2)
	90	self.c0 = nn.Conv2d(40, 40, 3, padding=1)
	91
	92	self.max_pooling = nn.MaxPool2d(kernel_size=2, stride=2)
	93
	94	self.c1 = nn.Conv2d(40, 60, 3, padding=2, dilation=2)
	95	self.c2 = nn.Conv2d(60, 40, 3, padding=2, dilation=2)
	96	self.c3 = nn.Conv2d(40, 20, 3, padding=2, dilation=2)
	97	self.c4 = nn.Conv2d(20, 10, 3, padding=2, dilation=2)
	98	self.output = nn.Conv2d(10, 1, 1)
	99
	100	def forward(self,x):
	101	x_red = self.max_pooling(F.relu(self.red_cnn(x), inplace=True))
	102	x_green = self.max_pooling(F.relu(self.green_cnn(x), inplace=True))
	103	x_blue = self.max_pooling(F.relu(self.blue_cnn(x), inplace=True))
	104
	105	x = torch.cat((x_red, x_green, x_blue), 1)
	106	x = F.relu(self.c0(x), inplace=True)
	107
	108	x = F.relu(self.c1(x), inplace=True)
	109
	110	x = F.relu(self.c2(x), inplace=True)
	111	x = self.max_pooling(x)
	112
	113	x = F.relu(self.c3(x), inplace=True)
	114	x = self.max_pooling(x)
	115
	116	x = F.relu(self.c4(x), inplace=True)
	117
	118	x = self.output(x)
	119	return x

Hints:
Before first commit, do not forget to setup your git environment:

git config --global user.name "your_name_here"
git config --global user.email "your@email_here"

Clone this repository using HTTP(S):

git clone https://rocketgit.com/user/hahattpro/crowd_counting_framework

Clone this repository using ssh (do not forget to upload a key first):

git clone ssh://rocketgit@ssh.rocketgit.com/user/hahattpro/crowd_counting_framework

Clone this repository using git:

git clone git://git.rocketgit.com/user/hahattpro/crowd_counting_framework

You are allowed to anonymously push to this repository.
This means that your pushed commits will automatically be transformed into a merge request:

... clone the repository ...
... make some changes and some commits ...
git push origin main