Wrecking Situationship

This is Karen. Karen loves gardening and collecting stuffs for decoration. Glass bottles (hard to find one without cracks), sea shells (her favorite), stuffed dinosaur (it looks like her long lost aunt Helen), and many more. There are some wrecked ships as well, but she claims "it was just a phase".

That is Bob. Bob is Karen's boyfriend (he may have not known this). Bob doesn't know anything about gardening, but he likes playing in Karen's garden. His favorite game is hide and seek, usually it's him hiding waiting to be found by Karen.

Last year they adopted two mantas together (Bob may also have not known this, but he likes them anyway). Karen named them Luca and Benjamin. Today, Bob took them out for a swim as usual, but it has passed dinner time and he is still not back yet. Karen is quite angry now. She starts looking for Bob.

Samples: 1024 spp
Resolution: 2160 x 3840
Render time: ~10 hours and 30 minutes

Features Implemented

This is a list of features that were implemented in the final rendering.

Alpha Masking

To test out our alpha masking implementation, we placed Bob under an opaque fabric. It also helps Karen to find Bob easier. Don't worry, no octopus was harmed in the making of this scene :)

Some other tests with alpha masking features:

Sheen Material

The stuffed dinosaur that Karen collected has a velvet-like material. The sheen material helps to achieve that soft and fuzzy look by simulating the way light interacts with tiny fibers on the surface.

Differences between 0.5, 0.5 and 1.0 sheen:

Normal Mapping

Normal mapping is an important feature that widely used in this rendering. In this example, the normal mapping recreates the cracks on the pillar surface without adding extra geometry, making the rendering more efficient. Many other objects in this scene also use normal mapping, such as the mantas and the wooden boat.

Some other tests with normal mapping features:

Image Denoising

Even with 1024 samples per pixel, some noise is still visible in the final rendering. Image denoising contributes greatly by smoothing out the noise while preserving important details.

Comparison between without and with image denoising:

Rough Dielectric

The glass bottles in Karen's collection have rough surfaces that giving see through effect with reflection and refraction. The rough dielectric material simulates this effect, making the bottles look more realistic.

Differences between 0.1, 0.3 and 0.5 roughness:

Spot Light

While a directional light can be used to simulate sunlight, we want to create a more focused light source that mimics sunlight passing through the gaps. The spot light allows us to achieve this effect by controlling the direction and spread of the light, resulting in a more soft faded lighting effect that add much depth to the scene.

Area Lights

On the other hand, the area light is used to simulate the soft light to brighten up the dark area. We combined different area lights color to create a more natural lighting effect that resembles the underwater environment while also ensure the image is not too dark. You can see there are two different shadow color on the statues: one is blueish from the water light, and the other is yellowish from the sandy ground reflection.

Comparison between without and with area light:

Other Features

This is a list of features that were implemented but weren't used in the final rendering.

Clearcoat Material

We didn't want to pollute the ocean floor with anymore plastic trash, so we decided to not use the clearcoat material in this scene.

Differences between 0.1, 0.5 and 1.0 coat:

Bloom

Considering our scene is underwater, the bloom effect may not be very suitable here. Thus, we didn't use it in our final rendering.

Comparison between without bloom and with bloom effect:

Low-Discrepancy Sampling

While the Halton sampling greatly reduces noise in the test, we already achieve a good quality of the final rendering with 1024 spp and image denoising. In this case, using Halton sampling may not bring significant improvement to the final image while adding extra complexity to the implementation. We decided to not implement it in the final redering for this particular reason.

Comparison between without halton sampling and with halton sampling:

Acknowledgements of Contribution

We would like to thank the following free resources for the models and textures used in this project:

Boat

Wooden boat model and texture by Kitaec312

Wrecking ship model and texture by GSXNet

Corals

Deep-sea coral (1) model and texture by MBARI

Deep-sea coral (2) model and texture by MBARI

Green coral (Pocillopora eydouxi) model and texture by thehydrous

Purple coral (Acropora) model and texture by yuma.jp

White coral model and texture by Nik

White pink coral (Anthozoa) model and texture by RISD Nature Lab

Kraken

Kraken model and texture by mrkblckwd

Landscape

Sand landscape model by aaakunkun

Wet sand texture by Poliigon

Objects

Glass bottle model by Blender3D

Stuffed dinosaur model by oraweb

Broken pillar model by Polygonized

Demeter statue model and texture by Bonn Center for Digital Humanities

Rocks

Rock model and texture by dimdimich

Boulder model and texture by lml46

Coral covered rock model and texture by MBARI

Sea animals

Clown fish model and texture by gimyujun392

Fish (1) model and texture by assetfactory

Fish (2) model and texture by MaxLeburo

Fish (3) model and texture by bukharig0300

Tillapia Buttikoferi fish model and texture by Ra_in_coat

Manta ray model and texture by misaooo

Octopus model and texture by QubiCone

Seashell

Seashell (1) model and texture by blenderboom

Seashell (2) model and texture by Meshfinder

Seashell (3) model and texture by mecha_ecl

Seaweed

Seaweed model and texture by grinvel

Starfish

Starfish (1) model and texture by QubiCone

Starfish (2) model and texture by Zygomir