vault backup: 2025-04-02 09:29:37

Affected files: .obsidian/plugins/text-extractor/cache/27f6eda968ec7bf3e9b3624aa45434a5.json .obsidian/workspace.json 0 Journal/0 Daily/2025-04-02.md Attachments/Pasted image 20250402091926.png
2025-04-02 09:29:37 +02:00
parent c4b1ef8142
commit 54c169210c
4 changed files with 33 additions and 5 deletions
--- a/.obsidian/plugins/text-extractor/cache/27f6eda968ec7bf3e9b3624aa45434a5.json
+++ b/.obsidian/plugins/text-extractor/cache/27f6eda968ec7bf3e9b3624aa45434a5.json
@@ -0,0 +1 @@
+{"path":"Attachments/Pasted image 20250402091926.png","text":"- INEW . Floris 18:04 ok. in the new algorithm i proposed, i think this distortion will be reduced if you chose a large closest_object_dist value. because then small gradients in the closest object (the stand in this example) will translate to a smaller gradient in the metric distance (edited) changing the disparity_constant is kindof dumb, of course the distortion becomes smaller if we reduce the camera shift (edited) it's like saying \"if we don't change the image, we don't have the problem\" n Nemanja Vujadinovié 19:31 | don’t understand why you have to call this dumb (and it’s not the first time), especially when | never once said that was the solution, nor did | say, “if we don’t change the image, we don't have the problem.” My point was that, since the disparity constant is currently just an experimental value, maybe it shouldn’t necessarily be 100—it could be lower. 100 is just an arbitrary value. On the other hand, yes, increasing the closest object distance will work because it limits the upper bound of the shift. If we set the closest object to 2 instead of 1, the disparity won't exceed 50, which is mathematically equivalent to simply setting a lower disparity constant. But logically, that doesn’t make sense to me, because it assumes there’s no object closer than 2 meters, which won't be true in many cases. . Floris 20:48 i called it \"kindof dumb\" because it is. (edited) G @ » R à : 20:51 it doesn't solve the problem reducing the disparity constant simply reduces the camera shift so then the diagonal distortion problem becomes less visible but it doesn't solve it and yes, the idea is equivalent to saying \"we don't have the problem if we don't change the image\" because that's what happens if you set the disparity constant to zero you can set the disparity constant to O or to 0.01 or something, and then the diagonal distortion problem becomes invisible because we're not shifting the camera i didn't mean to offend you btw i just wanted to point out a flaw in the approach n Nemanja Vujadinovié 21:03 And | never said that changing the disparity constant solves the problem. In the message and the image above the last one, | explained that due to inaccurate depth map there’s an apparent skeweness of the stand. Since upper and lower parts of stand differ in depth, they get shifted differently causing the skeweness. That causes the problem Moreover, in one of the replies Claudio said that disparity factor is too big and that stand shouldnt be shifted that much. | agree with that and that’s why | posted the results with different disparity factors/constants . Floris 21:11 ok i don't know why you take it so personally, i just wanted to point out that the disparity constant is not relevant to the problem. the problem is that there's a diagonal distortion sometimes, on foreground items (i've seen it in other pictures too). The fact that the disparity constant affects this distortion is trivial, because disparity constant is simply a scalar value that amplifies all the pixel shifts. but yes maybe i read too much into your message you weren't proposing it as a solution, you were just pointing out the visual effect of the distortion constant anyway, i do think that a higher closest_object_dist value solves the problem. the problem is this: the bottom of the stand has depth value O. The top of the stand has depth value 50. so if the range is 5 meters, and the closest_obj distance is 1m, that means that the algo will assume the bottom of the stand is 1m away, and the top of the stand is 2m away. that's why you get the diagonal distortion. on the other hand, if the closest_obj_dist is set to 3 meters, that means the algo will treat the bottom of the stand as being 3m away, and the top as being 4m away. this is what i meant when i said \"i think this distortion will be reduced if you chose a large closest_object_dist value. because then small gradients in the digital depth will translate to a smaller gradient in the metric distance\" (paraphrased) (edited) another potential diagnosis of the problem is that these (0-255) depth values are not linear maybe we have to take the square root or log or something","libVersion":"0.3.2","langs":"deu+eng+fra"}
				`@@ -0,0 +1 @@`
				{"path":"Attachments/Pasted image 20250402091926.png","text":"- INEW . Floris 18:04 ok. in the new algorithm i proposed, i think this distortion will be reduced if you chose a large closest_object_dist value. because then small gradients in the closest object (the stand in this example) will translate to a smaller gradient in the metric distance (edited) changing the disparity_constant is kindof dumb, of course the distortion becomes smaller if we reduce the camera shift (edited) it's like saying \"if we don't change the image, we don't have the problem\" n Nemanja Vujadinovié 19:31 \| don’t understand why you have to call this dumb (and it’s not the first time), especially when \| never once said that was the solution, nor did \| say, “if we don’t change the image, we don't have the problem.” My point was that, since the disparity constant is currently just an experimental value, maybe it shouldn’t necessarily be 100—it could be lower. 100 is just an arbitrary value. On the other hand, yes, increasing the closest object distance will work because it limits the upper bound of the shift. If we set the closest object to 2 instead of 1, the disparity won't exceed 50, which is mathematically equivalent to simply setting a lower disparity constant. But logically, that doesn’t make sense to me, because it assumes there’s no object closer than 2 meters, which won't be true in many cases. . Floris 20:48 i called it \"kindof dumb\" because it is. (edited) G @ » R à : 20:51 it doesn't solve the problem reducing the disparity constant simply reduces the camera shift so then the diagonal distortion problem becomes less visible but it doesn't solve it and yes, the idea is equivalent to saying \"we don't have the problem if we don't change the image\" because that's what happens if you set the disparity constant to zero you can set the disparity constant to O or to 0.01 or something, and then the diagonal distortion problem becomes invisible because we're not shifting the camera i didn't mean to offend you btw i just wanted to point out a flaw in the approach n Nemanja Vujadinovié 21:03 And \| never said that changing the disparity constant solves the problem. In the message and the image above the last one, \| explained that due to inaccurate depth map there’s an apparent skeweness of the stand. Since upper and lower parts of stand differ in depth, they get shifted differently causing the skeweness. That causes the problem Moreover, in one of the replies Claudio said that disparity factor is too big and that stand shouldnt be shifted that much. \| agree with that and that’s why \| posted the results with different disparity factors/constants . Floris 21:11 ok i don't know why you take it so personally, i just wanted to point out that the disparity constant is not relevant to the problem. the problem is that there's a diagonal distortion sometimes, on foreground items (i've seen it in other pictures too). The fact that the disparity constant affects this distortion is trivial, because disparity constant is simply a scalar value that amplifies all the pixel shifts. but yes maybe i read too much into your message you weren't proposing it as a solution, you were just pointing out the visual effect of the distortion constant anyway, i do think that a higher closest_object_dist value solves the problem. the problem is this: the bottom of the stand has depth value O. The top of the stand has depth value 50. so if the range is 5 meters, and the closest_obj distance is 1m, that means that the algo will assume the bottom of the stand is 1m away, and the top of the stand is 2m away. that's why you get the diagonal distortion. on the other hand, if the closest_obj_dist is set to 3 meters, that means the algo will treat the bottom of the stand as being 3m away, and the top as being 4m away. this is what i meant when i said \"i think this distortion will be reduced if you chose a large closest_object_dist value. because then small gradients in the digital depth will translate to a smaller gradient in the metric distance\" (paraphrased) (edited) another potential diagnosis of the problem is that these (0-255) depth values are not linear maybe we have to take the square root or log or something","libVersion":"0.3.2","langs":"deu+eng+fra"}