The Development of AstroEdit Planet 2.0

071126

I manage to get the stacking right now. 10 frames for 160 seconds. I am using search radius 48. think will reduce to 40 seconds when I used 0.25 size.

1. The micro shift is max about 36 and 44 pixel. so I have to use 48 as search radius.

2. The reserve mapping is used. it prevent black hole due to float offset. Some integer pixel may not get a color. To do reserve mapping, imagine a moon at x=100 in best frame. It has drift to x=130 at next frame. To start, we use the best frame coordinate, and loop through the pixel. the moon at x= 100, need to add 30 to get x=130 in the next frame. So the shift is next.x - best.x, or 

-> shiftX = next.x - best.x. 

And for shiftY = next.y - best.y. Since the frame center is tracked using Vision, it is bottom left coord for the frame center. To change to top left, we use next.y(top left) = 1-next.y. 

shiftY = next.y - best.y

shiftY = (1-next.y) - (1-best.y)

-> shiftY = best.y - next.y 

3. The coord top left applies to planar texture. The AP uses it too. Only the Vision input and output result rect are bottom left. So the best AP is y flip before send to vision. And the vision track frame xy is also bottom left coord. The

061126

manage to get rgb from biplane. the image is invert. so ap creation is on a invert image. on iPhone, 10 frame needs only 20 seconds. 

925pm I fail to stack 2 frame. what goes wrong? double image. max search radius.

041126

To speed up, I am working on biplane. it can score and track center for 130 frames in 6 seconds. It is fun. 

031126

It takes only 10 seconds for 10 frames on MacBook but takes 40 seconds on my iPhone. With the GPU fasten, it is still about 40 seconds on my iPhone 12 mini. 

I need to use astroplanet method to open and do frame score. it will reduce to 20 seconds. My ap is top left in astroplanet but bottom left in surface app. Do astroplanet open video first time for score. Second time for stacking?

Should I remove vision and use accumulative best ap matching? Should I remove mtltexture and use cifilter reduction? 

021126

I am doing 1 second for 1 frame for a 1/4 texture matching. 

1. The problem for max ap movement of 12px is due to using texture that is bottom left vs ap that is top left. Change ap to bottom left solves the problem. AI could not find the cause. It is when I read the code from the beginning that has texture. is texture also bottom left? yes.

2. Vision needs to update the observation in every request. I got it only when I study apple sample. I cannot use the initial rect which will be out of range at 120 frame. At 9 frame, it is ok. AI is wrong to say using initial rect is enough. The sample uses updated observation rect. I got the idea of using vision sample out of the blue.

3. The variance for frame on gpu does not work when 10000 is not enough, it needs 1000 000 to get a positive value. I get it when I am waiting for the orange juice machine at 3am. why do I use debug. ai and debug info, I get the 10e6 fix.

- do 1 time frame variance

- Integration with planet stacking. 

011126 

It is not the mipmap level 2 problem. The problem is the classic top left vs bottom left. I have changed to texture which uses bottom left. So besides vision, texture also uses bottom left. As such, I move the ap to use also bottom left. 

I manage to find the bug by reading the code again. AI is not able to find the bug. It could not see that texture is using bottom left. Android does not have this problem. In the beginning, I am using buffer so there is no texture bottom left issue. I change to texture because I want to use the level mipmap. 

- Add in the level 2 mipmap for fast execution. 

301026

Almost done all issues. time to integrate but find stack 19 become blur. Stack 9 is still ok. it is a reoccurring error. it is interesting to know what is the cause this time. It is interesting to write down on blogger. On 231026, mu record show I solve the problem by using baby vision step.

vision works well for 19 frames. the problem may be the mipmap 2. I will need to test no mipmap for 19 frames.

261026

Manage to reduce to 15 seconds for 10 frame using 4x reduction, astroplanet read, 10 frame gpu. but the ap creation has problem and the stack is blur. I need to work everyday to feel good. I have missed one day watching tang dynasty.

241026

I got so distress with the 29 frames blur. I start a baby vision project. it works. ask why main project fail, it turns out the vision rect need to vertically flip. It is baby step and AI that save the day. Now I need to quicken, check is surface and integrate to PlanetEdit.

231026

I get blur when stack 29 frames. I have to try vision baby step if it can track my whole video correctly. 

211026 Working on Moon surface since 6 days ago. It is as tough as planet. I need a good crater to track the global shift. I will need a good marketing effort.

051026 Upload version for approval. Finally. Takes 1 month 5 days for version 3! I nearly want to give up. It is so difficult that I have to start from basic view again.

300926 It is slow when I take baby step but at least I enjoy what I am doing and I can make progress. Taking giant step is fast but I got stuck at certain point. as I don't understand the code, I cannot progress. It is satisfying to see the graphics. Only thing is that I have to start from zero again to build 3 views. I have shift from AP defined at the corner to AP at the center. It is easier to draw with AP center but in matching will need to minus half ap size.

290926 when I slow down to the slowest, I enjoy coding again. Here is the right image that I understand how it works. metal bottom left or swiftUI is top left. 

 

280926 I find I have to restart again. It is getting so complicated with ap and stacking. I could not easily port the advance version to astroPlanet. I have to go as slow as possible to make it enjoyable.

180926 Push back as I find AP to work on with crop size at the edge of frame. It is tough to draw ap on view 1.

140926 manage to get all files working except ver. it is mainly the crop issue over frame size and 1-y_coord for metal data in view overlay. I have been working on AstroPlanet for a month!

130926 manage to get JupiterDob moving target working. It is important to ask the right question. 

100926

- in Penang, run at 6.9 for 60 frame sample jupiter.avi. can run on iPhone. I wonder how long to run a 30gb 8000 frame video. now going for test all video file, last will be ver. I am only working on iOS avi.

040926

Start improvement on speed and quality. the work is getting harder now.

010926 14 days to do.

- Done avi and ser file. rename avi to ser in order to copy to iPhone. Rename back once copied.

- work why stack 1% ser give blur result. stack again looks good.

- change is subscribe to Save button.

- Ask the right question and have a discussion to align the objective solve the integration of sub project ver, avi and mp4 into PlanetEdit.

- 32 frames gpu stacking.

310926

- working on ser, avi and mp4 sub project gpu quality sorting and viewing

300826

- able to load frame sequencely.

- do GPU open avi and ser

- check how long to read a 3g file

- add open to files and photo

290826

1. Use gpu to apply Warping too. 

2. Use gpu to analyse frame time for quality and AP

3. Use gpu to find polynomial coefficients. too complex for gpu?

4. best frame is not center in display tab hold.

5. Add frame to stack in lucky beside %.

6. change to 99c

7. One ser and avi file

270826

Double to reset all settings. Level to split the ring? Use 40% of screen height for controls. Fix load photo upside down. Add text Loading sample video... for inscriber. k

limit to 10 seconds of the file. Reading 10x30=300. It needs to get quality. Then it needs to get AP. Can they be combined? So it takes half the time? k

image link tp planetEdit in astroediy iOS.

260626 - Take 11 days to code. My longest is 5 weeks for cometEdit. First long is celestialPiano for 1 month. This time I only update the stacking and wavelet, the rest like UI and color are done in version 1 for about a week.

- take App Store screenshots, privacy link.

- test subscription. new code but with user default for faster loading.

- test wavelet, color and open photo, video.

To do: - Need to add image to astroedit next. 

250826

- Ingrate big project (AP, bounding box and stacking) into on sale project(existing PlanetEdit) takes some hours with a break. when ask the right questions it works. I skip my social to do this, it takes me much longer than I think.

240826

- test jupiter, juoiterdob, saturn, saturn2. got bounding problem with jupitermov. will do frame baby test next. look like binding box is the crucial obstacle to overcome! can use AI? It is exactly caused by a hot pixel. Median works. I have tested for seestar30, seestar50. they give no visual detail improvement. I have test on jupiterzwo. 

- I am going to integrate to my main program. Do I need to set lucky image or just set as 5% since it gives the best result? Maybe keep it as others may get good 20%. 5-20 range.

- Integrate big project to astroplanet. it has the stacking, opening video, astroedit wavelet, a smooth slider, rgb alignment, color and subscribe button (will need to test the subscribe and restore). $0.99 a month. Add combine 4 legacy desktop window software into 1. PIPP to crop, AutoStakkert! to stack, Registax wavelet and photoshop to color. do it conveniently on your iPad. remove stack alert to text update. display top 1 frame if lucky is 0.

230826

- test jupiterdob is ok, fix some AP not on planet. Test saturn2, the blue bounding box is not on Saturn, will try again on baby bounding box.

220826 

Manage to get AP grid on the bounding box without user select them on Saturn. Using threshold of 75 is better than using iOS vision contour to find the bounding box. filling bounding box full of AP is better than only placing AP at the Saturn. I will be adding 20 frames with next button. If ok, do all 20 frames at once. manage to do all 20 frames automatically. Should I hide all complexity in stacking and just output the result with text stack 1...20?

Manage to do stack from Saturn.mp4. Will test 10 video files. there are plenty on Facebook now!

How to avoid if the Saturn is corrupted in 1 frame, maybe someone knock the telescope. 

210826

I was playing with affine transform to stack the frame18 to frame1 in a baby project. It creates double ring even with Ransac. Then it comes to my mind, why not use the gradient removal. And it works like abc. It links back to the time I think of graph used to model a woman curve. And now after 52-25= 27 years, the dot links by looking backward. Without the idle time on the bus 382 ride, I would not have the idea to use math to create woman breast.  And it never comes to me that it will be useful 27 years later. 

200826

frame 10 onwards it starts to fail. getting to 0/9 which is impossible. at least 3 points will hit. these images tends to move more, elongated and blending.  double ring still occurs.

look like I have to go into 2 frame baby step!

190826

The result is still blur with top 10%. What's wrong? I plan to use GPU to do the AP feature matching next. I wish I could see why it stacks blue at the ring. Which frame causes it? I used 20 frames. Maybe I should print out the frame number. I wish I could see the original frame with AP. Then the problematic frame with its AP. I could then see the cause for the blurriness. 

Maybe a 4 views, First show top frame with AP. Second show next frame with search AP. Third show top frame with red AP. On top overlay the next frame (transformed) with AP. Fourth show the accumulated stacked frames. Then there is the next and back button. so I can scroll until the stack get blurred.

Manage to get gpu to work. now the matching is fast. Got the 4 views, my double ring blur is caused by 3 ap near the lower left. the 3 ap points are all way of of the best frame ap. why they does not match? beside the best frame is brighter. is it stretched?  

180816

I work on the matrix solver. Using a small project with google AI, I manage to get the transform matrix right. Previously I am struggling with it with the project for hours and still not working. 

170816

I manage to get to use Ransac to choose a good transform that satisfy at least 5 points. I have used 9 AP. 

160816

Start a swiftUI project asking google AI for Saturn stacking. It uses translation for stacking. I ask to add in 3 points affine transformation. The result is still blur with top 20%. What's wrong? 

150826

I start to work on a video cropper app to crop a Saturn video to around its size. Alternatively, I could use the original video and do real time cropping and stacking. If so, I don't have to track the global translation of the Saturn. It is always centred. 

In August, I notice that Earth is coming to Saturn and Jupiter in October using my AstroWidget app. I See more people will video these giant planets and I decide to work on AstroEdit Planet 2.0. I am recording my development so that someone can continue my work in the future.