Classifying Novel 3D-Printed Objects without Retraining: Towards Post-Production Automation in Additive Manufacturing

Abstract

Reliable classification of 3D-printed objects is essential for automating post-production workflows in industrial additive manufacturing. Despite extensive automation in other stages of the printing pipeline, this task still relies heavily on manual inspection, as the set of objects to be classified can change daily, making frequent model retraining impractical. Automating the identification step is therefore critical for improving operational efficiency. A vision model that could classify any set of objects by utilizing their corresponding CAD models and avoiding retraining would be highly beneficial in this setting. To enable systematic evaluation of vision models on this task, we introduce ThingiPrint, a new publicly available dataset that pairs CAD models with real photographs of their 3D-printed counterparts. Using ThingiPrint, we benchmark a range of existing vision models on the task of 3D-printed object classification. We additionally show that contrastive fine-tuning with a rotation-invariant objective allows effective prototype-based classification of previously unseen 3D-printed objects. By relying solely on the available CAD models, this avoids the need for retraining when new objects are introduced. Experiments show that this approach outperforms standard pretrained baselines, suggesting improved generalization and practical relevance for real-world use.

Figures (12)

• Figure 1: ThingiPrint Dataset. Examples of paired CAD models and real images of their 3D-printed counterparts. Our dataset includes 100 CAD models sourced from the Thingi10K dataset, each accompanied by multiple real-world photographs of the corresponding 3D-printed objects.
• Figure 2: Examples of objects in ThingiPrint. Our dataset consists of various 3D-printed objects with diverse geometries.
• Figure 3: Example of the data collection process using smart glasses. The user positions the object within the rectangular region displayed on the device, and this region is cropped to obtain the final image.
• Figure 4: Object capture. Example photos of a single object from our dataset. The object is hand-held and rotated to capture a variety of orientations.
• Figure 5: Overview of the classification pipeline at inference. (Left): At test time, we render images for each CAD model and compute a prototype vector $p_i$ to represent each object by averaging the representations of the encoded images. (Right): During inference, we classify an image of a 3D-printed object by comparing its feature representation $z_t$ to the computed prototype vectors and picking the object ID $o$ with the highest similarity.