The codes of Cloud-Robotic AI Benchmarking for Edge-cloud Collaborative Lifelong Learning #65
Conversation
Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn>
|
@luosiqi might want to take a look at it |
| @@ -26,7 +26,7 @@ cd /ianvs #One might use another path preferred | |||
|
|
|||
| mkdir project | |||
| cd project | |||
| git clone https://github.com/kubeedge/ianvs.git | |||
| git clone -b feature-lifelong-n https://github.com/kubeedge/ianvs.git | |||
There was a problem hiding this comment.
In the formal version, the clone shall be revised to the main branch
| if mode != 'multi-inference': | ||
|
|
||
| # in this mode, the inference period is skipped to accelerate training speed | ||
| if mode == 'no-inference': |
There was a problem hiding this comment.
Is this code for test not for formal running? If it is, please modify it.
There was a problem hiding this comment.
This code provide multiple modes for running, users could choose either test or formal running in "algorithm.yaml".
| inference_dataset = self.dataset.load_data(self.dataset.test_url, "eval", | ||
| feature_process=_data_feature_process) | ||
| kwargs = {} | ||
| #model_eval_info = self.model_eval_config |
There was a problem hiding this comment.
Annotations should be removed.
There was a problem hiding this comment.
Ok, I will remove annotations.
| for i, _ in enumerate(inference_dataset.x): | ||
| #print(i) |
There was a problem hiding this comment.
Annotations should be deleted.
There was a problem hiding this comment.
Ok, I will remove annotations.
| file.writelines( | ||
| f"{os.path.abspath(os.path.join(root, front))} " | ||
| f"{os.path.abspath(os.path.join(root, back))}") | ||
| #front, back = line.split(" ") |
There was a problem hiding this comment.
Annotations should be removed.
There was a problem hiding this comment.
Ok, I will remove annotations.
| @@ -148,6 +173,12 @@ def split_dataset(self, dataset_url, dataset_format, ratio, method="default", | |||
| data_types=dataset_types, | |||
| output_dir=output_dir, | |||
| times=times) | |||
| # add new splitting method for semantic segmantation | |||
| if method == "my_splitting": | |||
| return self._my_splitting(dataset_url, dataset_format, ratio, | |||
There was a problem hiding this comment.
The function name "_my_splitting" is not explicit.
There was a problem hiding this comment.
Ok, I will change the name of the function
| else: | ||
| test_loader, num_class = None, cityscapes.CityscapesSegmentation.NUM_CLASSES | ||
|
|
||
| # custom_set = cityscapes.CityscapesSegmentation(args, split='custom_resize') |
There was a problem hiding this comment.
Annotations should be removed.
There was a problem hiding this comment.
Ok, I will remove annotations.
| return len(self.images[self.split]) | ||
|
|
||
| def __getitem__(self, index): | ||
| #print(type(self.images[self.split][index])) |
There was a problem hiding this comment.
Annotations should be removed.
There was a problem hiding this comment.
Ok, I will remove annotations.
|
|
||
| return model | ||
|
|
||
| # def main(): |
There was a problem hiding this comment.
Annotations should be removed.
There was a problem hiding this comment.
Ok, I will remove annotations.
|
I suggest that the commits can be compressed by using "git rebase -i". |
|
As you have modified core codes of ianvs, make sure it works for other paradigms such as single task learning and incremental learning. |
hsj576
force-pushed
the
feature-lifelong-semantic_segmentation
branch
from
edcefce to
c96951a
Compare
OK, I have compressed the commits. |
I am pretty sure that the modified codes will not influence other paradigms. |
|
It seems that the CI testing is not passed. |
|
Please modify your codes according to CI. It seems that your codes don't meet the coding standard. |
I have modified the codes to meet the coding standard. |
Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> add example for semantic segmentation Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> add example for semantic segmentation Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> fix some bug Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> fix the bug of memory leakage Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> fix the bug of inference Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> fix some bug Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> fix the bug of test accuracy Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> edit sedna package for lifelong learning Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> edit sedna package Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> implement dataset split method for lifelong learning Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> add test examples Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn>
Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> write quick start Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> remote useless file Signed-off-by: hsj576 <sjhu21@m.fudan.edu.cn> revise the git command Signed-off-by: husj <hushijing19990403@126.com> remove annotations and change the implisit function name Signed-off-by: husj <hushijing19990403@126.com> modify the code to meet the coding standard Signed-off-by: husj <hushijing19990403@126.com> modify the code to meet the coding standard Signed-off-by: husj <hushijing19990403@126.com> modify the code to meet the coding standard Signed-off-by: husj <hushijing19990403@126.com>
hsj576
force-pushed
the
feature-lifelong-semantic_segmentation
branch
from
5b79d54 to
a74a843
Compare
|
/lgtm |
What type of PR is this?
This PR provides codes for PR#62
What this PR does / why we need it:
It provides a benchmark example of semantic segmentation for lifelong learning in Ianvs.