Week Six (Film Room Evaluation)

The sixth week of this term marks the completion of our 221b Film Room project. Over the course of this period we have grown much more familar with uses and applications of PBR and the Unreal Engine, as well as strengthening the ability to work in a team environment. Below you will find my 2000 word evaluation of the creative process leading up to the hand-in. Some sections may be edited versions of earlier blog posts if I found that they provide the information required by the brief.

Pre-Production

Our very first task was to look into which environments to consider. We were relatively well decided that the chosen environment ought to be complex enough to be visually interesting, but not so much as to jeopardize the time scale with too much work. The three primary film scenes that we considered to have a good number of such traits were Bilbo/Frodo's house in The Lord of the Rings, the Burrow from the Harry Potter saga and the Brethren Court Pirates of the Caribbean.

We ruled out the LotR entry when I pointed out that the interior architecture of the ceilings and walls were all very arced and curved, meaning that replicating it to an accurate extent in-engine would likely use up a very large chunk of our alloted triangles. Although we liked the dark and brooding lighting, the Brethren Court was scrapped for two reasons: the first being that we failed to find any high resolution images of the scene (and what we could find was all from one angle), and the other being that most members of the group had not seen the films and would therefor not be very familar with it. On a side note, the sheer number of actors in the scene shown in the image above would have made it harder still to gage what the room looked like without parts obscructed frrom view. The strongest suggestion, the Weasley's Burrow was seriously considered, as it had appealing lighting and plenty going on, but was eventually discarded due to the sheer overwhelming number of items that in the scene that we would have had to produce

After some more searching, we decided to choose the 221b set from the BBC's television show, Sherlock. On the left you will see our chosen shot taken from one of the episodes, and on the right, my edited version. As you can see here, there's more than enough objects to keep us interested, and the drawn curtains provide a strong, diagonal shaft of white light which illuminates the dust particles in the air and falls softly in the very centre of the room, picking up the worn, bobbly texture of the rug below. As well as this sunlight, there are also a few forms of artificial light emanating from the various lamps and picture frames, which ensure that the illumation is spread relatively evenly throughout the room. This made not only for a good composition, but also plenty of potential for our PBR textures to catch the light. There are also a wide range of textures in this shot - plenty of woods, canvas materials, leathers and metals which will likely evoke interest in the eyes of the viewer.

The Process

Because I had chosen to do 2 of the largest items in the entire scene - the walls and floor, I realised at one point during the texturing process that my textures themselves were translating to the models quite poorly, heavily muddied and low resolution. This was because I had unwrapped the entirety of all four walls onto a single 2048x2048 pixel canvas, which although it is on the higher end as far as texture maps go, wasn't anywhere near big enough for the diffuse maps to appear a good enough quality. In hindsight, I had considered that this might be a problem whilst I was engaged in the unwrapping stage, but I didn't know how to address it and simply carried on, which was probably a mistake on my part. I likely could have saved some time by talking to my fellow team members about possible solutions a little earlier, but once I was told that it was ok to use sub-materials, I pretty much got on it straight away as I was quite excited to see my models in all their high resolution glory. In the end, converted a lot of the textures to HD tileable versions and assigned them to a material ID in 3DS Max, which worked just fine.

Above left you will see some of the progress I made during the first week of the project - an integral part of the reference shot, the wall with a book base on each side. I decided that I would construct all the main architecture as one singular model rather than doing the bookcases or remaining walls separately to avoid any nasty looking overlaps or gaps. This decision generally worked in my favour, although I made a critical error in the bookcase shelve's topology quite early on, resulting in many unwanted overlapping vertexes meaning I had to start again. That said, I was quite confident that the walls I had replicated from the show were accurately proportioned - you may notice the translucent plane peeking from being the model in the first image, which was the scaled reference pane I had used to make sure the bookcases and general structure were just right. I achieved a decent front-on view of this particular wall by finding an on-set photo and manipulating the perspective, which I then used along with the measurements reverse-engineered from Sherlock's height to determine an approximate height and width of the room.

In the upper half of this image you can see 3 UV maps which I produced before I decided to use sub-materials. As you may or may not know, using this technique means that each material ID is tiled infinitely over the canvas - not just the central square for UV unwraps. This meant that I could place the UV's anywhere I needed, which not only gave me a lot more space, but also allowed tiling textures to appear at a much higher resolution. Showing you the final UV's here wouldn't be practical, but the ones you can see here are pretty much the identical, just all packed into a neat square.

I produced a total of 10 textures, some tileable, some more specific. Difficulty finding appropriate "starting point" images for each texture varied significantly during the process, for example, wallpaper 1 is was very easy to find on Google, but wallpapers 2 and 3 were not. This was likely due to the fact that wallpaper 1's design had caught more of the viewers eyes whilst watching the show (I in fact noticed that it is quite commonly used as a background for Sherlock fan sites) and has therefor gained popularity, becoming known as the general "Sherlock wallpaper". In the end, I had to make textures for both of these designs from scratch, using photo reference. For wallpaper 2, I extracted an example of the tiling pattern from a high-res promotional photo, played around with the perspective to make it appear as if directly facing the camera, and touched it up into a much larger and crisper seamless texture.

When the time came to import all the assets into the engine, there was a short period where everything had to be shuffled around, rotated and resized relative to my walls, as they were the closest match in scale to the real-life version. After everything had been positioned appropriately, we hit another barrier - as of that point, we hadn't actually positioned the camera to match the reference shot, meaning that we also had to adjust the field of view and reposition the pilot camera around the room, something which should really have been done whilst we assembled the original whitebox mockup. Mike Pickton mentioned that we should have addressed this at a much earlier stage so that when we replaced the boxes with actual assets, very little adjustment would need to be done, which is something we will definitely need to make time for next time. Importing these models into Unreal also made another previously invisible error glaringly apparent - in the right-most image you may notice that the engine's default checkerboard texture is horribly distorted! This was due to a mistake I had made with the positioning and scale of the window whilst unwrapping, and thanks to Tara for pointing out, I was able to quickly rectify it and re-import. Angelos actively sought to alert us if any remodelling needed to be done, too - at one point I had to adjust the height of the overbearing ceiling above Anthony's shelf units to a more suitable value, moving it from just below the door's level to just above.

My Project Performance

Personally, I'm happy with how my assets turned out, especially since this was my first time using Unreal and Bitmap2Material. I had more confidence in some models more than others, but I believe that the texturing process really pulled everything together, as I wasn't really sure how my stuff would look until it was fully textured. The modern armchair is a prime example of this mindset; after completing the model I still thought that it looked polygonal, it needed a more natural and lived-in look. My PBR did just that, as the tutors say, "A good model can't save a bad texture, but a good texture can save a bad model" (not that I consider my armchair to be a bad model, just a little too smoothed over and artificial).

Looking back, I would have probably just asked for more assets to do at the beginning. For a while we thought we only had 5 weeks to get this project done, and then it turned out to be 6. We were on track to get the scene done for the fifth week and this news slowed us down - we managed to stretch it out another week. Although this was relaxing to hear, if I'd known earlier I could probably have got more done. Also, though I had my fair share of assets, others had much more complex shapes to model, and though you could argue that I had my armchair, I would have liked to get a bit more creative. The walls were challenging at times, with all the skirting boards and attachments, but there's no denying that they're flat and angular - hey, they're walls! They were always going to end up having a lower tri-count than some of the other assets but I believe I could probably have gone a little further and taken on more things.

Overall, I can safely say that the final result exceeds my previous expectations. Looking at some of the other groups' work had me feeling a little concerned at first, as few of them were ahead of us in the production process and had better looking stuff to show off. But now we've applied our textures, PBR and fancy lighting I think our film room definately deserves to be held just as high as that of the other groups!

Team Manifesto

When our group came together for the first time to decide our film room, we were surprisingly easy to negotiate ideas with. As the usual group project starts we shared ideas for possible rooms, which included iconic sets such as ‘The Burrow’ from the Harry Potter film series and ‘The Hobbit House’ from the Lord of the Rings Trilogy. There were also a few less known sets like the laboratory from Time Machine (2002).

However, it was the BBC’s TV adaption of Sherlock that finally agreed with every member of our group. We felt that the number of assets in our chosen shot (as shown above) as well as the overall mood of the lighting would work well in our favour. As we all knew of or had watched the TV series, we knew that the room/set had a good balance of large and small assets. This meant that when we divided up assets every person got to model and texture a main piece of furniture and the clutter surrounding it. This proved to be useful in dividing up leftover assets.

Divvying up the assets was one of the easiest steps to the development of the project as we all had particulars that we wanted to do, and although there were certain assets that more than one person held interest in, we were able to compromise efficiently. In the end, all six of our team members got either a piece of comfort furniture (sofa, armchair, etc.) or a bookcase with content. Something that was also important to us was making sure that each team member got an asset in view of our chosen still. This was not difficult as Sherlock’s apartment is ridiculously full of content.

An issue we came across in gathering reference for our individual assets was that annoyingly, the Sherlock set changes in every episode. So what paraphernalia was in one episode may not be in another or at the opposite end of the room. The assets that are clearly visible in the chosen still were easy enough to replicate but the team members who got assets in the unseen section of the room had to choose an appropriate episode to draw reference from.

Throughout the project we all worked at different paces, but we gave clear deadlines weekly as to where we should be as a group. As an example we set a group deadline to have modelling finished by end of Week 3 and for the most part we succeeded in that.

During our process we all came together in class to import our room in UE4 to see its progress and decide the next course of action. This was something we tried to do on a regular basis until we made the final UE4 file that would become our hand-in scene.