Ttingyyy
commented
5 months ago Open Ttingyyy opened 5 months ago
Ttingyyy
commented
5 months ago 
TQTQliu
commented
5 months ago Hi, these black points can be considered as noise points, which will affect the accuracy metric. These noise points can be filtered out by increasing the thresholds (confidence and geometric thresholds) at the fusion stage.
Or you can change here, in our implementation, the confidence map of the last stage is applied for fusion. Another way is to multiply the confidence maps of all stages together as the final confidence map, like:
photometric_confidence = confidence_list[0]*confidence_list[1]*confidence_list[2]*confidence_list[3]Of course you need to adjust the confidence threshold accordingly. The latter is a more rigorous filtering method that produces point clouds with fewer noisy points.
KAWHIPAUL
commented
3 months ago Hi, these black points can be considered as noise points, which will affect the accuracy metric. These noise points can be filtered out by increasing the thresholds (confidence and geometric thresholds) at the fusion stage.
Or you can change here, in our implementation, the confidence map of the last stage is applied for fusion. Another way is to multiply the confidence maps of all stages together as the final confidence map, like:
photometric_confidence = confidence_list[0]*confidence_list[1]*confidence_list[2]*confidence_list[3]Of course you need to adjust the confidence threshold accordingly. The latter is a more rigorous filtering method that produces point clouds with fewer noisy points. Hello, first of all, thank you for your excellent work. I have encountered the same problem. I have a question, is it still possible to achieve the results in the paper after the modifications you made in your previous answer. Looking forward to your answer.
TQTQliu
commented
3 months ago @KAWHIPAUL
The strict filtering method can obtain better Acc., while the loose filtering method can obtain better Comp.. However, the metric Overall is the average of Acc. and Comp.. I did not try to carefully adjust the hyperparameters of the strict filtering method, so I do not know whether the Overall result in the paper can be achieved after modification to strict filtering.
05063112lcs
commented
3 months ago Hope to receive your reply! I have some doubts about the point cloud results obtained from testing: Why are there some black point clouds around the obtained point cloud?
Hello, I want to ask, you asked before that there is noise around your point cloud, you solved this problem, if it is solved, can you please tell me the specific operation, thank you very much
KAWHIPAUL
commented
3 months ago Hope to receive your reply! I have some doubts about the point cloud results obtained from testing: Why are there some black point clouds around the obtained point cloud?
Hello, I want to ask, you asked before that there is noise around your point cloud, you solved this problem, if it is solved, can you please tell me the specific operation, thank you very much Sorry, I can't get the metrics in the paper and a good visualization at the same time. According to the author's answer, I changed the conf to 0.2, and there will be a lot fewer black dots, but the indicators will be worse. Hope it helps.
05063112lcs
commented
3 months ago After you change the conf index to 0.2, can the visualization effect correspond to the article? After you modify the indicators, the relevant parts in the test_dypcddtu.py file have been modified. Photometric confidence = confidence list [0] * confidence list [1] confidence _ list [2] confidence _ list [3]. After I revised the index to 0.2, the index is very close to the article. I suggest you try it several times. Are you still doing the follow-up research on this article?
---Original--- From: "Ye @.> Date: Tue, Jul 2, 2024 11:12 AM To: @.>; Cc: @.**@.>; Subject: Re: [TQTQliu/ET-MVSNet] test：Why are there some black point clouds around the obtained point cloud? (Issue #12)
Hope to receive your reply! I have some doubts about the point cloud results obtained from testing: Why are there some black point clouds around the obtained point cloud?
Hello, I want to ask, you asked before that there is noise around your point cloud, you solved this problem, if it is solved, can you please tell me the specific operation, thank you very much Sorry, I can't get the metrics in the paper and a good visualization at the same time. According to the author's answer, I changed the conf to 0.2, and there will be a lot fewer black dots, but the indicators will be worse. Hope it helps.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>
KAWHIPAUL
commented
3 months ago After you change the conf index to 0.2, can the visualization effect correspond to the article? After you modify the indicators, the relevant parts in the test_dypcddtu.py file have been modified. Photometric confidence = confidence list [0] * confidence list [1] confidence _ list [2] confidence _ list [3]. After I revised the index to 0.2, the index is very close to the article. I suggest you try it several times. Are you still doing the follow-up research on this article? … ---Original--- From: "Ye @.> Date: Tue, Jul 2, 2024 11:12 AM To: @.>; Cc: @.**@.>; Subject: Re: [TQTQliu/ET-MVSNet] test：Why are there some black point clouds around the obtained point cloud? (Issue #12) Hope to receive your reply! I have some doubts about the point cloud results obtained from testing: Why are there some black point clouds around the obtained point cloud? Hello, I want to ask, you asked before that there is noise around your point cloud, you solved this problem, if it is solved, can you please tell me the specific operation, thank you very much Sorry, I can't get the metrics in the paper and a good visualization at the same time. According to the author's answer, I changed the conf to 0.2, and there will be a lot fewer black dots, but the indicators will be worse. Hope it helps. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>
Thank you for your answer. I am still continuing to study this issue. Can you provide me with a contact information for easy communication?
05063112lcs
commented
3 months ago After you change the conf index to 0.2, can the visualization effect correspond to the article? After you modify the indicators, the relevant parts in the test_dypcddtu.py file have been modified. Photometric confidence = confidence list [0] * confidence list [1] confidence _ list [2] confidence list [3]. After I revised the index to 0.2, the index is very close to the article. I suggest you try it several times. Are you still doing the follow-up research on this article? … ---Original--- From: "Ye @.> Date: Tue, Jul 2, 2024 11:12 AM To: _@_.>; Cc: @._@_.>; Subject: Re: [TQTQliu/ET-MVSNet] test：Why are there some black point clouds around the obtained point cloud? (Issue #12) Hope to receive your reply! I have some doubts about the point cloud results obtained from testing: Why are there some black point clouds around the obtained point cloud? Hello, I want to ask, you asked before that there is noise around your point cloud, you solved this problem, if it is solved, can you please tell me the specific operation, thank you very much Sorry, I can't get the metrics in the paper and a good visualization at the same time. According to the author's answer, I changed the conf to 0.2, and there will be a lot fewer black dots, but the indicators will be worse. Hope it helps. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @_.***>
Thank you for your answer. I am still continuing to study this issue. Can you provide me with a contact information for easy communication?
OKOK，my wechat 18993571134
Hope to receive your reply! I have some doubts about the point cloud results obtained from testing: Why are there some black point clouds around the obtained point cloud?