Danial Thompson – Rotoscope Assignment 01

1.0 – Finding the shot

I underestimated the importance of shot compositions and did not fully understanding the ensuing implications that could arise from choosing a difficult shot. I initially found eight shots of varying quality and setting. I was keen to follow a theme involving walking-assistance apparatus. I presented examples to my tutor, Allar Klaasik, and whilst he agreed that the shots were interesting he informed me that they were missing technical elements.

The issue across all of my shots was that there wasn’t a great deal of natural motion blur. According to Allar, the ability to rotoscope motion blur accurately would be an important and impressive addition to our showreels (Klaasik:2017). This was, in essence, my first setback as the struggle to find appropriate footage ate into my schedule. I ended up settling between a shot of a formula car and a shot of two workers. Allar gave them both the green light. I consulted my class tutor Klaudija Cermak for her opinion too. She preferred the formula car. Klaudija also suggested that if I were to do 100 frames of the car, I could do a short 50 frames of my original high-fidelity walking apparatus. It was my intention to follow this advice and indeed it wasn’t until much later into the project that I realised two shots would not be feasible.

1.1 – The Formula Shot

The Formula Shot is a 16 second, 1920 x 1080 H264 recorded at 25 frames per second. Out of those 410 frames, I selected a range of 100 between frame 69 and 169. I selected this range because I felt that the car was in good focus and there was good interaction between the organic human element and the inorganic car in the opening of the shot.

The fast opening moves of the organic human creates a lot of motion blur. I am inclined to believe that, whilst part of this blur is natural, there must be some element of operator control contributing to the overall ‘blur’ of the of the movements. Usually, you want the denominator of your shutter speed to be approximately double the number of frames per second that you are recording. I think that the cameraman was using a slow shutter speed hence the overall softness of the shot. It did mean that going forward my motion blur wasn’t very selective, it was uniformly applied to the vast majority of shapes to account for the shutter speed.

In terms of motion, there were many elements operating independently of each other. There was no definable pattern to the motion paths of focus objects and camera moves (Bratt:2011). The racecar moves on a seemingly gentle A to B path, which is quite deceptive. In fact, there is a lot of jitter and motion on the car. This is in part due to the bumps in the track surface, but also the material of the car itself. The carbon fibre body reacts to every move, particularly the wing which wobbles precariously. If the right wheel went over a bump, the right framework would shake, but the effect of that bump might not travel across to the left side of the car. This is something I began to notice as I looked at the car at the pixel level.

The inorganic elements are also tricky. The ‘Friend’ who is the leading character in the foreground does a three-point turn. This is a motion that I didn’t really pay much attention to until I had to include him in the shot. It would be a source of struggle later on. The background character ‘Mechanic’ moves in a fairly linear horizontal fashion. The camera pans to the left towards the end of the shot.

2.0 – The Racecar

I decided to tackle the largest object on the screen first. On a personal note, I enjoy motorsports so I was keen to start working on the car. I perhaps naively thought that it would also be the easiest element. Further showing my inexperience, I thought I had a ‘leisurely’ amount of time and I did not tackle the shot with any great urgency initially.

2.1 – Tracker novice

My opening move was to find some way to stabilize the shot to eliminate movement of the camera. The racecar presented many locations on which to track so stabilizing was not an issue. I then applied shapes under stabilization. I did not consider at this time that my tracker was on the left of the car and that whilst I had removed the camera motion the shapes I was making were not following the bumps in the road. I was not letting the software help me interpolate bumps, particularly on the Y-axis.

2.2 – Sub-object manipulation

In my haste, I had not considered any form of reasoned approach. The car presented many definable areas but initially, I just wanted progress. As a result of not utilising trackers effectively, I found myself keyframing a lot of bumps, particularly on the wing. This indulgence led to the tampering of the sub-object level. On this path of self-destruction, I reached a critical turning point in the assignment. Having received Benjamin Bratt’s book ‘Rotoscoping: Tools and Techniques for the Aspiring Artist’ I flicked through a few pages. I quickly realised I was making mistakes, and his particular words:

“If you continue working inefficiently, you’re not going to improve.  Not fixing those habits might have a more dire effect: You might actually become slower.”

2.3 – Backtrack

I had scrapped my shapes a total of two times. I realised how inefficient this process of discovery was. I considered that reading the topic and watching a few tutorials before attempts wouldn’t hurt. I took the trade-off of losing a day or two to this preparation over the estimated time saved changing the workflow. Bratt’s book, in particular, gave me a good grounding with regards to how to tackle my shot. It reflected in the speed at which I started to tackle the shot. I felt more confident with my judgements knowing that I had a mental checklist to tick off as I went.

2.4 – Bifurcation and Tracking

I began attempting to track the racecar compartmentally. Initially, I created too many trackers that seemed to be working counter to each other, throwing my shapes off.

tracker off - Danial Thompson.png

One example of tracker application gone wrong is when I tried to apply one to the wing layer within my car folder, but after doing that all the other shapes shifted. When I deleted the wing tracker in an attempt to revert to the original stable tracker, the entire shape began to move differently to how I wanted.

Later I discovered that it was probably a result of not applying the track on the right keyframe. The initial keyframe where the shapes are created establishes the scale for that shape at 0%. If I applied trackers in the middle of my timeline, I can believe that it would do strange things to the shot.

too_big_shape - Danial Thompson.png

I started with a big shape on the wing, but because the tremor at the top of the wing was much greater than at the base I decided to sub-divide the edges to account for the shake. I also added a new tracker focused specifically on the wing. This was the process of discovery for many parts of the racecar.

I was mindful of bifurcation as I knew that it is much smoother than what I could accomplish (Wright:2017). In my eagerness, I had neglected this process. I had to begin going back to my focus edge shapes and refining the keyframes – something I should’ve done from the outset. I also noticed inaccuracies and once I noticed it all the shapes along that focus edge seemed in need of adjustment.

I was also experimenting with matchmove, which worked well with the inorganic. The jitter looked more natural and the shapes required little in the way of tending. I was annoyed that earlier I didn’t consider trying harder to make the matchmove trackers work before keyframing manually. Worse still, I felt like the areas I had done purely through bifurcation were now lacking in comparison to the matchmoved areas.

low_keys - Danial Thompson

I was getting carried away with segmenting areas of the car. I decided upon observation of the rear that it would be cleaner to have just one group for that region and one match-move tracker instead of more layers in the object list. With more shapes came more opportunity for crashes. At one point, a two of my shapes had disappeared from my viewer, even though they were in the object list. I couldn’t figure it out even with tutor help. It was a reminder that you always run some form of risk using the latest versions of the software.

remove_rear_Left - Danial Thompson.png

noshape - Danial Thompson.png

3.0 – Presentation Feedback

With a week to go, Allar and the tutors reviewed our progress thus far.

By this stage, the car was complete. The alpha looked convincing and the colour comp was good. I had not blurred any of my edges yet so it looked sharp, and there were a few shapes moving out of sync. Apart from that, it was positive. The ominous observation was that I had elected to rotoscope the mechanic (of whom I had only done a foot) when actually It makes more sense to do the two characters, foreground and background. This made sense and I naively believed that the foreground ‘friend’ would not be too much work compared to the motion-heavy mechanic.

4.0 Mechanic

I found it incredibly difficult to get a track on the Mechanics legs. They moved so quickly and the shutter speed was so low that the tracker couldn’t keep up. To further compound this, foreground Friend casts a shadow over him mid-way through the motion so the tracker loses any point assigned. To begin with, I was manually keyframing shapes from frame 69 right through to 169. I realised however that the second half of the lower right leg motion is a lot simpler than the first action. Subsequently, I manually keyframed the shapes to the still moment just before the second move at frame 102. From 102 onwards I group moved the rest of the leg, which was overall quicker than guiding shapes across 100 frames individually.

frame1-2 - Danial Thompson

I struggled with the jeans clumping. There were many jagged shapes which really needed individual shapes. As a result, when it came to keyframing those forwards it became difficult to retain a sense of cohesion. I tried this with the 102 frame stop method, but 102 to 169 was really difficult and the end result looked quite chaotic. I was not convinced I had chosen the correct method at this point.

As I moved to the upper leg, I resolved to track it. I had to do it manually and I was worried about accuracy. It did, however, help me to translate the shapes from A to B and also allow for some stabilisation. This worked well. At the time, I thought to employ it from then on for each area. I believed that if the process was sped up enough there would be time at the end for me to go back and revisit the lower right leg and apply a similar method. At the time I was forced to press on.

jeanage - Danial Thompson.png

The general over-exposure of the shot and the strange compression of detail meant that when the two characters criss-crossed each other’s denim it became hard to distinguish Mechanic from Friend. Unfortunately for the shot not even gamma blasting seemed to help. I had some luck isolating individual colour channels, however.

Late one evening I felt the project slipping away from me. Particularly because at each step I was so critical of my work and whether or not I was doing it right I forgot the big picture. I turned on all the shapes and looked over the whole roto thus far and realised it wasn’t looking too bad.  This helped overcome blocks and walls on more than one occasion.

5.0 The Friend

For a long time, I remained ignorant of the effect this character would have on the shot. He was, in fact, the anchor I needed to cement my shot.

5.1 Abandoning the Mechanic

I got stressed out by shapes passing behind the foreground Friend and seemed unable to keep shapes coherent behind him. I decided to get him blocked out first instead to understand how that Alpha would react with shapes passing behind it and give me a space to prepare and adjusts shapes for smoother transitions.

hmmm1 - Danial Thompsonthe explosion - Danial Thompson

As I looked at all my shapes one evening in Alpha channel, I was not happy at all with the front-right leg and what happened when it crossed behind Mechanic. There was a lot of bouncing and it didn’t seem like a solid object at the time. I attempted to see whether Mocha tracker would find a way where I could not, just out of curiosity. I ended up going on a journey to the next dimension and quickly realised it wasn’t working.

interstellar___ - Danial Thompson.png
I’m coming back Murph

Remembering to be mindful of key bunching was difficult. As the deadline loomed, I was too pre-occupied with complex shapes. Keeping on top of the timeline keys became a secondary concern – which I knew was bad. Compared to the relative composure of my inorganic timeline, there was very little logic to my organic timelines. The reality of rushing was dawning on me. Rushing keyframes was particularly worrisome for the Friend, as for a good 50 frames he remains quite still, making any bubbling very obvious.

6.0 – Integration of Mechanic and Friend

phase 1 - Danial Thompsonphase two - Danial Thompson

edit these photos to annotate what you mean

As I was doing the Mechanic I reduced shapes to 0% opacity. The shapes then went behind the Friend matte. It was only later when I realised that the body re-appears again when Friend rotates. Thankfully, the Mechanic remains stationary during those frames. I realised the friend acts like an eraser, giving me a clean slate with each pass.

I decided to do the end-phase of the Mechanic as a separate set of shapes. There is very limited movement so the Mechanic is a lot crisper. I was dubious that I could use the shapes from when the Mechanic was in motion as a lot of detail was lost to the blur and shutter. This detail would need to be there when he was stationary. I realised that the Mechanic would be useless as a standalone alpha as a lot of DIY patching happened ‘behind the friend’. This was again a product of running out of time.

Example of phase one isolation from phase two (with the exception being the legs)

I realised that even though I was cutting corners by making shapes disappear, it would’ve been nicer if shapes had at least disappeared more uniformly.

QC_CHECK_BLUR

As I neared the deadline I began QC checks to make sure shapes had uniform blur. This was important because shapes appeared in and out with opacity. I should’ve just selected all the appropriate layers in the object list and set the blur. Instead I had ad-hoc applied blur to some shapes and not others. I had to manually scrub through to pick out sharp edges. Depending on their position of entry, if they are not blurred it can look very noticeable on the alpha.

I had the idea that I should apply more blur to the Mechanic in order to emulate focal depth. I decided against this because upon closer review, when stationary the figures share similar focus detail.

7.0 – Layer Organisation

When organising layers, I wanted to make it so that they were layered in descending order and by group. I also noticed that I didn’t really have a uniform naming convention. Sometimes I would have a location and then specific location (Upper_left / left_upper) this struck me as odd. I decided to make it uniform with the PART of the body preceding all other details.

Screenshot-10 - Danial Thompson

Thus I tried the naming convention:

body part / location (upper, lower) / fill /edge / fold / L/R / #

Until present, I was using only the location and body part. I considered that there are so many of the same location and parts of the two bodies that perhaps being more specific is appropriate. I devised this naming convention in the context of if I were giving the project to someone and they wanted to find a shape using the search tool. Whilst I thought it was a good idea I ended up confusing myself. As the nights wore on I started getting the layer naming back to front. Time pressed on and I found it difficult to prioritise layer management with shapes still to finish.

too many colours - Danial Thompson.png

This is the result of failing to organise on the go. Many shapes that could belong to upper or lower. I first sorted them into their position and then mass coloured them to either an upper colour or lower colour for clarity before starting the renaming process. Renaming and organising was a huge time drain. Mass splines and layers accounted for four Silhouette crashes. I couldn’t work out why initially but quickly realised that moving too many layers and shapes into a new layer can cause Silhouette 6.1.4 to seize up.

8.0 – Conclusion

To conclude, I found my first rotoscoping assignment very challenging. There was a lot to take in and more than one setback that made me appreciate the thorough nature of rotoscoping. Discipline and a logical plan of attack are clearly what’s needed to achieve quality results. Though it isn’t perfect, I already feel much better equipped to tackle the next shot.

In hindsight, I would have prioritised my elements differently. The organic elements caused me the most trouble and were the ones that needed time to be accurate. I would have created the matte for Friend straight away and use that as a pillar. Knowing what I know now about the strategic use of trackers, the racecar would have been less daunting to finish. I would’ve liked to have been more organised and to have tidied up my keyframes. I am overall unhappy with the organic timelines.

I need to learn how to remain cool under fire. I felt the pressure of time and neglected practices that I should be enforcing diligently. On the note of time, I would not allow it to get so ahead of me. My estimations of rotoscoping were wildly off, leading to a large amount of detailed work crammed in the last 50 hours. As a result, all the quality checks I wished to do were not done. I only have myself to blame in that regard.

Regardless, I enjoyed the experience and am proud of my attempt. I have learnt a lot.

References

Bratt, B. (2011). Rotoscoping. Burlington, MA: Focal Press., p. 70 -128

WRIGHT, S. (2017). COMPOSITING VISUAL EFFECTS. [S.l.]: CRC PRESS, p.22.