The week 4 we have studied deeper functions of 3D camera tracking. The first interesting new method to remove some specific patches is the node of the project. I froze a certain frame to roto paint for removing some useless patches. Then I froze the same frame again to hold the result of roto paint again.
Now the question is that the patches can’t follow the camera to move. So I project the image to a grid that catches the same position with tracking cloud information. Besides, I need to freeze the same frame as before, otherwise, the image will shift in the grid by each frame.
Then I use the original camera to render the scene again, I can get the patch to catch the camera moving. What I did like to extract an image to fix to a wall and use the original moving camera to render again.
But this method has a major disadvantage. At the same time as the first froze the frame, it also froze the shadows. This method cannot be used when there are large variations in light and shade. So I learned another method.
We use the node of the project to replace the node of the frame hold. Stabilisation of the picture with little distortion helps us to finish roto painting.
This week we learned some new things about 3d Tracker which is similar to the 3D equalizer.
The main workflow and principle are the same with other tacker software, however when I want to re-distort the image the side of the image will have some errors such as the below image.
this week We have learned about constraint and animation. Because keying frame is a little complex, I just create a circle as a controller and using script to links each element together . If I want to change speed at any frames I can only changing one parameter to control all things. Then I used some noise function to control the speed of wheels.
We also learned something about mash. I am also interested in the workflow of the mash UI, however I hardly ever used the mash part in maya, so I will cost more time to know more about these.
This week we almost finished all the methods to remove markers. I will collect all methods by my own understanding and I won’t post some simple way that does not get a great result.
The first method I want to call the High and low-frequency method because this way of the principle is very similar to the PS. The way separate images into two parts by dividing node. One of them records the low-frequency information without detail only having colours and another one has high-frequency detail. Then only changing the high-frequency detail and combining them, we can get a nice result. There is a more advanced skill for fixing the high-frequency detail. I can transfer the original images a little to get similar detail instead of roto paint.
The second way is Roto paint which uses a patch directly to cover the marker. We have an advanced method to track each pixel, smart vector node. The node is different from the track node. the smart vector can analyze each pixel vector information in different channels for each frame. Then the VectorDistort node will transfer raw data of the vector to use by other nodes.
The third way is to fill pixels by unpremult node.
Except the third method is a little different principle from others, the main aim of methods is to cover the marker by other pixels in a specific frame, then track the position to cover the marker in the whole sequence. We use a different way to improve each workflow for saving time and reduce work pressure to avoid painting each frame.
Homework link
This sequence is not easy to track even I tried my best to adjust the colour to extract more contrast. Such as the picture below, the left line I want to divide the picture into two parts and one of the parts will keep colour, another one keep other information. Then I use rotopaint node to cover the marker shape and copy the marker transform position to the rotopaint node. However, the result is not perfect, because the specific marker in cheekbones has obviously changed of shadow and shape deformation. I combine another method to solve certain frames and mix them in end.
For the other 5 markers, I found that if I adjust all tracker positions and each error in different frames, will cost too much time. So I directly use a simple workflow to remove all markers like below. Although there are some node is different with what I learned in class, the principle is the same. I use rotoPaint3 to cover other markers frame by frame.
For adding a texture to the sequence, I directly use a texture and copy tracking information to it, then it works which can attach the face to move.
For the machine example of Mr Nick Savy, I noticed that he added a key to the model UV offset to “cheat” eyes. However, in this way, a UV map is required to unfold to a specific type otherwise, we will get a wired result. Besides, he made a blender shape for the belt and the tiny shaking will make the belt look like working normally.
For myself, I really like to make a steam machine such as the reference images.
Based on the old machine, I will add a belt to it. Now I almost finished half of the model. Such as below:
The steam will push the stick to support power for the wheel. Then I want to add another wheel to change the kinetic energy direction.
