Cottage

The mill attendant’s cottage was the primary focus of the 2015 excavation of the site undertaken by Heritage Victoria. It was built in 1855 to house the attendant of the nearby mill, constructed at the same time (Smith et al. 2016: 82). The cottage is partially visible in an 1878 photograph (See Smith et al. 2016: 893), and features a brick chimney, wooden walls, and a veranda common in early colonial Australian architecture. Period appropriate comparanda was consulted in the reconstruction process (Keep In press).

As only the bluestone foundations, remnants of a brick veranda floor, and brick chimney foundations of the cottage remained in-situ, the model was created entirely within Blender without using any photogrammetric modelling. The dimensions of the cottage were deduced from the floor plan provided in the excavation report (Smith et al. 2016: 84) and the elevation of the roof was estimated on the basis of comparanda and the existing historical photograph (see Figure 2). The main body and roof of the cottage was modelled as a single object with multiple textures applied to different sections, while individual elements such as the veranda beams and windows were modelled individually and duplicated as necessary.

Physically based rendering (PBR) textures were used to allow image textures to modify the geometry of the structure procedurally to reduce time spent modelling and the processing load of the Blender scene. A subdivision surface modifier was used to add vertices to otherwise flat and featureless surfaces with a displacement image used to modify the geometry according to data derived from the PBR textures. This workflow created a more realistic, varied, and ‘textural’ surface appearance without excessive modelling and rendering time investment. The same process was used to add texture to bricks, wood beams, wood slats, and roof tiles. The resulting complexity transformed otherwise flat and lifeless surfaces into a more realistic result (Figure 3).

Other Structures

In addition to the cottage, the mill and nearby aqueduct were modelled to create a more immersive interconnected scene rather than an isolated cottage. These structures are less thoroughly excavated than the cottage. An 1878 etching by Samuel Calvert (Figure 4) depicting both structures served as the primary reference for reconstruction, while relevant comparanda provided the details, in place of primary archaeological data. Historical photographs of the nearby mills provided comparanda for design, and there were sufficient archaeological remains to determine that the structure was built primarily in bluestone with a brick chimney (Vines 2017: 525), similar to other mills of the region and period. The mill and aqueduct were, like the cottage, modelled using polygonal modelling and rendered with PBR textures using a displacement modifier (Figure 5).

Marble

A single hand painted ceramic marble decorated with concentric circles painted in red, green, and purple-brown was recovered from the site during excavation (Tucker et al. 2016: 47). This find was considered of high significance in the interpretation of the site, as its decoration indicates that it was neither a bottle stop nor used in blind baking, but was specifically made as a children’s toy. This indicates that the site was not only at least partially domestic, but was a family dwelling with a child present at some point during its occupation.

The marble was modelled using both photogrammetric modelling of the original artefact (Figure 6) and using PBR modelling from reference images within Blender (Figure 7). Digital photography using a stationary tripod-mounted Nikon D5300 was conducted inside a light-box using a bluetooth connected Foldio 360 turntable for semi-automated photography. An X-Rite ColorChecker Passport card was photographed alongside the object, with the associated colour calibration software used to maintain colour fidelity in Adobe Lightroom Classic 11.5, as recommended by Gaiani et al. (2017). Minor adjustments were made to the images within Lightroom (Table 1) to reduce the impact of glare and provide more texture for Agisoft Metashape Standard 1.8.2 (Agisoft 2022) to identify. As complete photographic coverage of the marble was impossible with the underside obscured as the marble rested on a surface, the model was produced in two sections (termed ‘chunks’ in Agisoft Metashape) which were subsequently merged together. The automatic process provided by Agisoft Metashape created a misalignment, and so manual alignment by eye was necessary before the halves were merged. The merged model was retextured using the complete image set for both halves of the marble. The division line between the halves is almost imperceptible, and would certainly not be noticed on the scale at which the model will be viewed (Figure 6).

The hypothetical reconstructions of the marble were modelled by hand in Blender, then imported into a scene featuring a dirt circle with modelled grass in imitation of the game of marbles. These models could be animated to collide with one another, to demonstrate in a short animation their function as a toy. This animation was rendered within Blender and incorporated as a video in presentations (Figure 7).

Pipe Bowl

Considered the most notable find recovered from the site by the excavators (Tucker et al. 2016: 52), a clay pipe bowl moulded in the shape of an imperial soldier was found in the mill excavation. Clay pipe use was widespread across 19^th century colonial Australia, and often acted as a signifier of class (Gojak and Stewart 1999: 40). The presence of this pipe on the site not only indicates the working class identity of the inhabitants on the site, but also suggests by its design a continuing cultural affiliation with imperial Britain by its owner.

An attempt was made to model the pipe within Blender before access to the original was available for photogrammetric modelling, but after several attempts the level of detail on the face and helmet proved beyond the author’s current ability as a 3D modeller. The face in particular proved exceptionally challenging. Likely a result of the human brain’s highly tuned capacity to recognize minute differences between faces, any model produced by hand would have to be of high quality to stand-in for the original, and it was decided that only photogrammetric modelling or a hired professional 3D artist would be capable of completing the task to a sufficiently life-like standard.

Access to the original was eventually possible after COVID lockdown restrictions had been eased. Photography was completed as above using a turntable and lightbox with colour correction, and minor adjustments in Lightroom (Table 1). As with the marble, complete photographic coverage was impossible, and the model had to be completed in two merged sections. Modelling the pipe-bowl interior presented a particular challenge, and bright lighting was used in an attempt to have this section modelled successfully. The interior walls were successfully modelled by Agisoft Metashape, however the lowest sections of the bowl were insufficiently lit, resulting in improper geometry and digital artefacts. Rather than leave this section as is, or obscure it from view, it was decided instead to recreate an estimate of the original interior geometry within Blender. A simple half-sphere mesh was joined to the distorted interior, and the texture from other sections of the pipe bowl walls was applied. Additionally, it was noticed while in Blender that the two halves were not evenly lit, with the upper half brighter than the lower: a result of the bright lighting used in an attempt to make the interior visible. To compensate for this, the texture for the lower half produced by Agisoft Metashape was adjusted within Blender for brightness and contrast, until it matched the texture produced for the upper half (Figure 8). The edited model was exported as a Graphics Language Transmission Format Binary file (.glb) so that these changes to texture would be embedded into the model itself. This model is currently available on Heritage Victoria’s SketchFab page (Heritage Victoria 2022).

Perry Davis Painkiller

A single bottle of Perry Davis’s Vegetable Painkiller, identified by the embossed glass text, was uncovered during excavation. Perry Davis’s Painkiller was an American opiate serum marketed internationally from the 1840s onwards (Tucker et al. 2016: 53), part of a larger market of unregulated curative formulas concocted and distributed in the late 19th century (Petty 2019: 264) period of laissez-faire industrial capitalism. The find is indicative of the connection of the occupants of the rural site to wider international trade networks, as well as providing further evidence for a working-class labourer identity for the occupant of the cottage as Perry Davis’ Painkiller was heavily marketed in Australia and New Zealand to workers such as miners and labourers (Petty 2019: 264) as the ‘trusted friend of the mechanic, farmer, [and] sailor’ (The New Zealand Herald 18 October 1881: 7).

The bottle was found complete and in a relatively good state of preservation, although the cork and label had long since been lost. Glass is a notoriously difficult material to model using photogrammetry alone, as the translucency and reflectiveness wreak havoc on image alignment. These issues can be surmounted through the use of scanning sprays or combining photogrammetry with X-ray CT scanning (Fried et al. 2020). Due to the delicacy of the opalised patina on the surface of the glass, the sprays available were considered inappropriate, and access to a CT scanner was not available. However, the patination of the glass did substantially reduce its translucency, and so an attempt was made to model the bottle using a standard photogrammetry workflow. Low lighting was used to reduce the impact of glare, with a longer exposure time to compensate (Table 1). Minor adjustments were made within Lightroom (Table 1), and the model was completed in multiple chunks which were successfully combined using Agisoft Metashape’s automated alignment. The final model presented better than anticipated, however there were inevitable warpings, aberrations, and digital artefacts that distorted the model. The material settings for the render in Blender were mixed with a combination of the image texture generated in Agisoft Metashape and a transparent BSDF (bidirectional scattering distribution function) node in Blender to reduce the opacity of the model and provide a greater sense of its original appearance.

Given these predicted difficulties in producing a photogrammetry model of the glass bottle and the reality of the tarnished patinated surface and lost labels, cork, and internal fluid, it was decided to create a 3D reconstruction within Blender depicting how the bottle may originally have appeared. Reference images of better preserved bottles were sourced online, and compared with the example from the site to ensure the correct design was used. From here, a replica was modelled within Blender, using the wireframe view to check proportions (the same process seen in Figure 2). Thickness was added to the 3D modelled bottle, and a volume was added internally to represent the medicinal concoction itself. PBR texture images sourced online were used for the cork material, while the materials for the glass and internal fluid were created using Blender’s internal ‘Principled BSDF’ shader. The labels were added to the bottle as flat planes, which were textured using images sourced online of well preserved examples. Text geometry was added within Blender using the ‘add thickness’ modifier to create the raised sections on the bottle which read ‘DAVIS’, ‘VEGETABLE’, and ‘PAINKILLER’. The final result depicts a highly life-like representation of the original bottle with cork, labels, fluid, and untarnished glass (Figure 9).

While neither the photogrammetry model nor the polygonal modelled hypothetical reconstruction are perfectly accurate representations of the original, when presented together they can provide audiences with an understanding of the conceptual link between artefacts and originals, revealing the process of archaeological inference while also building a sense of realism in the overall site reconstruction. A version of the reconstruction model is available on the author’s personal SketchFab page (Keep 2022c).

Red Gums

The red gums of the region have had a long and significant history and their presence continues to play an important role in the cultural landscape. They had an important pre-contact role in Wurundjeri resource exploitation, providing bark for use in the production of shields and canoes (Ellender 1991: 9–11; Griffin et al. 2013: 62; Cumpston 2020: 18). In the early colonial period European settlers exploited red gums for timber, such that timber frames became standard for pioneer structures of the period (Marsden 1989). Within the City of Whittlesea, the presence of dense forests spurred the development of a timber industry which supported growing settlement (Honman et al. 2013: 31–33), particularly important during the 1850s gold rush where the timber was used for posts, rails, and palings (Honman et al. 2013: 19). The surrounding terrain of the Plenty Gorge Parklands is still spotted with these trees. The longstanding cultural significance and prevalence of this species made it a priority for inclusion in the virtual reality reconstruction.

The reg gums were created in Blender using procedural modelling and PBR texturing. The base model of the trunks and branches were generated using the ‘Sapling Tree Gen’ add-on which follows the method developed by Weber and Penn (1995). This add-on automatically generates a base tree model with branches, and provides options to modify the branch count, number of divisions, trunk height, and width, etc. Using red gum reference images from Google, the base model was modified until it resembled the general form of the species. A set of leaves and gumnuts were individually modelled within Blender from reference images of samples collected from nature and photographed against a white background. With these reference images imported into the background of Blender, a plane was extruded and edited to match the shape of the leaves (Figure 10). The same process was repeated for the twigs and gumnuts.

PBR texturing relies on a combination of images to inform rendering software of the necessary interactions of light and geometry (Iakushkin et al. 2018). These image include scatter/diffuse maps which describe the colour of the surface, normal maps which describe the surface relief, height maps that describe the deviation of the surface, ambient occlusion maps that describe shadows created by surface detail, and specular or metallic maps that describe the intensity of light reflected from the surface (Iakushkin et al. 2018: 464–465). Thousands of these PBR textures are available for thousands of different surfaces freely or inexpensively on the internet, however for highly specific surfaces for niche applications, such as the appearance of a gum leaf, they can be difficult or impossible to find. It was therefore necessary to generate my own PBR texture maps (Figure 11).

This was completed using ‘Materialize’, an open source tool produced by Bounding Box Software which generates the PBR textures from imported images while allowing users to modify and adjust the generated textures (Figures 12 and 13). These textures were applied to the modelled leaves, which were then duplicated and modified to provide a variety of alternatives. The leaves were manually placed along a modelled twig, which was itself duplicated and modified. This process resulted in a selection of distinct branches from the base assets, so that duplicated models would not be noticeable, providing a more natural, organic appearance (Figure 14).

With the twigs completed, it was then necessary to texture the tree itself. Initially an attempt was made to generate a bark texture from photogrammetric modelling of a section of a red gum visited by the author in Mernda (Figure 15), however it quickly became apparent that the resulting model was far too complex and polygon-heavy for practical use without extensive optimization. Additionally, red gum bark and skin varies in appearance across different sections of the tree, so multiple photogrammetric models would need to be created and blended across different sections of the model. Although textures generated from a photogrammetry model would provide a highly detailed and realistic appearance, at the distance that the trees are visible much of this detail is lost, so the trade-off was made to instead use simple image textures.

Images of red gum trunks were sourced from Wikimedia Commons and uploaded to IMGonline to automatically align them into a ‘seamless’ pattern that tiles the image, using automatic processing to detect and minimise sharp transitions between the tiles. This image was downloaded and imported into Adobe Photoshop Classic 11.5 where the clone stamp and smudge tools were used to obfuscate noticeable seams that were not resolved by IMGonline’s processing. The final texture exported has minimal noticeable seams, allowing it to be wrapped, stretched, and repeated as necessary across the tree trunks without clearly detectable repetitions and transitions which would appear clearly unnatural and disrupt the immersive experience of the viewer (Figure 16).

With the texture applied to the modelled trunks and the branches created, it was necessary to combine them. This was completed using Blender’s internal particle system, which allows for objects to be procedurally emitted from selected surfaces (Álvarez-Cedillo and Almanza-Nieto 2010). Blender’s weight paint tool was used to manually paint sections of the branches that should have leaves and branches emitting, using reference images to guide the relative amount appropriate for different sections of the tree. This process was repeated with alterations using the same textures and assets to create a variety of red gum models (Figure 17). This workflow allows for the creation of realistic appearing trees with varied branches and leaves which interact with the lighting to cast realistic shadows and reflections (Figure 18).

Other Vegetation

Although not having the same cultural significance to Mernda as red gums, wattle bushes are also native to the region (Robinson 2020; Ellender 1991: 13). Wattle was used as a resource by Wurundjeri groups in the creation of axes and the production of resin adhesives (Griffin et al. 2013: 60–61), and was used by colonial settlers in the wattle-and-daub construction technique (The State Library of New South Wales 2009). Manna gums are likewise present in the Mernda landscape, representing the dominant species in dense forests on the Silurian sandstone terrain. By the period of the reconstruction much of the remaining landscape had been cleared of native vegetation for pasturing and bushfire control, with introduced grass species overtaking native murnong and grassland. Reeds and rushes remained prevalent along the waterline (Ellender 1991: 8–14).

These plant forms were created using the same process as the red gums (Figure 19), with collected samples photographed; photographs used to create PBR texture maps; trunks created using the Sapling Tree Gen add-on; leaves, flowers, and gumdrops modelled, textured and applied to the trunks using a particle system. Grass blades were modelled individually using the same process as tree leaves, and a randomiser modifier was used to collate them into clumps that could be placed over the landscape.

These natural elements were imported into a Blender file containing a digital elevation model (DEM) of the region which had been converted into shapefile (SHP) using the software package QGIS Desktop 3.12.3 (QGIS 2022). The landscape was weight painted with individual particle systems for the manna gums, grass, and red gums following the landscape units identified and mapped by Ellender (1991: 12). Low polygon duplicates of the manna gums without leaves attached were placed behind the standard models in areas where the canopy was no longer visible and grass placement was restricted to areas visible to the camera to reduce processing strain. The landscape itself was textured with dirt, marsh, grass, and understory PBR textures applied using a mosaic rotation and noise mapping node (Barber 2022). A reflectively shaded plane modified with ripples was used for the Plenty River water. With the terrain complete (Figure 20), the buildings and objects could be placed appropriately on the DEM derived landscape and cameras placed to render out stereoscopic images.