Questions about FCOS3D and PGD model 3D box #1024
Comments
|
|
Thank you very much for your busy response. I'll retest as you say. I will give you timely feedback on the results. |
|
Hello author： |
|
|
I see. I'll try again. Thank you very much! I will timely feedback the results. |
|
Hello author： And as you can see, the detection accuracy is also very low. What are the reasons for these two situations? What's the solution? For the second question, I conducted the test according to the format you provided. Unfortunately, there is still no way to save the result file. Here's an example: Error source code: def show(self, results, out_dir, show=True, pipeline=None):
"""Results visualization.
Args:
results (list[dict]): List of bounding boxes results.
out_dir (str): Output directory of visualization result.
show (bool): Visualize the results online.
pipeline (list[dict], optional): raw data loading for showing.
Default: None.
"""
assert out_dir is not None, 'Expect out_dir, got none.'
pipeline = self._get_pipeline(pipeline)
for i, result in enumerate(results):
if 'img_bbox' in result.keys():
result = result['img_bbox']
data_info = self.data_infos[i]
img_path = data_info['file_name']
file_name = osp.split(img_path)[-1].split('.')[0]
img, img_metas = self._extract_data(i, pipeline,
['img', 'img_metas'])
# need to transpose channel to first dim
img = img.numpy().transpose(1, 2, 0)
gt_bboxes = self.get_ann_info(i)['gt_bboxes_3d']
pred_bboxes = result['boxes_3d']
show_multi_modality_result(
img,
gt_bboxes,
pred_bboxes,
img_metas['cam2img'],
out_dir,
file_name,
box_mode='camera',
show=show)The source code is in the The same data key as variable For comparison, I have attached the same contents of the 000001 image from the I tried a lot of things. I hope I can try my best to solve the problem myself before bothering you. But my capacity is limited. I'm really hope you can help me solve the above problems. |
|
Follow the comments below. Similarly, when I tested the SMOKE model, I also had an error similar to problem 2. This seems to be a common bug. The following is an error message: |
|
It's really strange. Please check your prepared data and training log for the reason of the much lower performance than expected. You can take our provided logs for reference. |
|
I didn't do training. It was tested directly using the weighting model you provided. I prepared the dataset by strictly following the tutorial in the documentation, downloaded the full KITTI dataset, organized the format as required, and generated the training and test files. As for the low accuracy of the test I think it is caused by the fact that I only left 20 images in the validation set. I will rerun the test with the full validation set and submit test feedback. Until then, could you please arrange a colleague to see how to fix the bugs I've reported. Thank you very much! |
|
This is the result of a test using the full validation set. I annotated |
|
The error should be related to the compatibility of the function |
|
Thanks for your hard work! Looking forward to hearing from you guys! |
|
Hi! @Gewaihir , In PGD model, the score for each object is combined with object's uncertainty. If you print the scores: So you will find that they are not typically in range [0,1]. I recommend you to set a higher score threshold while visualization. |
|
@Gewaihir Another supplemental note is that a lower score threshold can empirically yield a little better (very little that can be ignored) quantitative results. So when I did the ablation I did not set it to a higher one. As for the compatibility of visualization API, we will fix it ASAP. |
|
Thank you both so much for your work! I will follow the suggestions and give it a try. Looking forward to the perfect code! |
|
Hello Author! I set the corresponding |
|
@Gewaihir You need to set the param |
|
@Tai-Wang I tried your suggestion and set different and even extreme |
|
@Gewaihir Did you change score threshold in pgd_r101_caffe_fpn_gn-head_3x4_4x_kitti-mono3d.py config? And for another problem about saving image results, a PR provided by us will fix the problems soon. |
|
@Gewaihir We suggest you to change the code like #1066 for a quick solution, or wait until the code is merged into our codebase. After code changing, the command for saving results should be like this: Actually, the problem is caused by inference model on multi-gpu testing scheme, Saving results while inference is only applicable to single GPU testing and mainly used for debugging and simple visualization. |
|
@ZCMax I changed the parameter of The entire terminal output is as follows. There are no saved files in the specified folder except for the error report. |
|
|
@Gewaihir the PR has been merged into our v1.0.0.dev0 branch, you can update your code via |
|
@ZCMax To avoid the problem in 2 again. I double checked and modified the code. It still didn't save any files after re-running. The new error reported is as follows. I added the following code at the beginning of the file that reports the error where the module is imported. import tensorflow as tf
tf.enable_eager_execution() Re-run test.py. The terminal displays the following message. Although no error was reported, it did not solve the problem. |
|
Hello @ZCMax I git pull the latest branch. Re-running the test resulted in the following error. The code file for this error is located in the |
|
@Gewaihir It seems you git clone a new repo, change from |
|
Results Show. Prediction Chart Three files are generated by inference. The file names are |
gt.png represents the image with ground truth label. |
|
@ZCMax Understood. Thank you！ |
Dear author：
Hello，Thank you very much for the framework and model you published！
I trained the FCOS3D algorithm using KITTI data set as described in issue #865. The profile is the same as issue #865, training 72 epochs. After the training, I tested the trained weight model and PGD weight model respectively, and encountered two problems:
1、I ran into this on both models: too many 3D boxes：
FCOS3D：
PGD：
As you can see. They all detect too many 3D boxes. PGD even has a bunch of 3D boxes that shouldn't be there. What causes this problem? How to solve it?
2、I wrote test instructions to test according to the official documents and git instructions. Unfortunately, it can only be viewed and saved one by one. Batch tests cannot be completed, saved, and then viewed. Is there any solution to this problem? Maybe I'm not doing it right? I enclose mine here：
python tools/test.py configs/fcos3d/fcos3d_r101_caffe_fpn_gn-head_dcn_2x8_1x_kitti-mono3d.py /media/lee/DATA/project/FCOS3D/latest.pth --show --show-dir /media/lee/DATA/project/dataset/FCOS
python tools/test.py configs/pgd/pgd_r101_caffe_fpn_gn-head_3x4_4x_kitti-mono3d.py checkpoints/pgd_r101_caffe_fpn_gn-head_3x4_4x_kitti-mono3d_20211022_102608-8a97533b.pth --show --show-dir /media/lee/DATA/project/dataset/PGD
I look forward to your early reply! Thank you very much.
The text was updated successfully, but these errors were encountered: