Visual Defects Research Articles

Cosmetic outcomes of cranioplasty using individual titanium implants

The cosmetic outcome of cranioplasty performed with individual implants is an important component in assessing the results of surgical reconstructions of scull defects. Developing temporalis muscle atrophy affects the harmonic face contour and, in cases of using patient-specific implants (PSI), designed taking into account bone symmetry, may discredit this technique due to unsatisfactory cosmetic results. In this regard, an urgent task is to find the possibility of compensating for a visual defect by modifying the shape of individual implants in order to achieve good cosmetic results. 54 cases of performed cranioplasty were divided into two groups: in the first group (n=32), the implant was modeled symmetrically to the intact side of the skull; in the second group (n=22), the curvature of the plates was modified to compensate for the developing temporalis muscle atrophy. To standardize the zone potentially responsible for the temporal muscle atrophy, a refining marking of the KrÖnlein-Bryusova scheme was carried out. To determine the optimal structural element that increases the strength of the plates in the area of their free edge, a number of load experiments were carried out in a specialized computer environment. Good cosmetic efficacy for periods of at least 6 months after the operation using symmetrical implants and localization of defects in the temporal region was noted in 68.7 % of cases, while with the use of modified implants this figure amounted to 100 %. The results of digital modeling have established that the rounding of the free edge is the optimal structural element that increases the strength of the plates. The developed approaches have a high potential for implementation in neurosurgical practice when performing cranioplasty in patients with defects of the skull bones localized in the temporal rectangle zone.

On-the-fly image-level oversampling for imbalanced datasets of manufacturing defects

Visual defect recognition and its manufacturing applications have been an upcoming topic in recent AI research. Defect datasets are often severely imbalanced and can be additionally burdened with separating classes of high visual similarity. Although various methods of data augmentation have been proposed to mitigate the class imbalance, they often fail to cope with tinier minority classes or have fidelity issues with smaller defects while, at the same time, needing significant computational resources to train. Also, augmentation based on vector-based oversampling struggles to produce high-fidelity inputs and is hard to apply on custom CNN architectures, which often perform better for this type of problem. Our work presents an image-level oversampling method based on an instance-based image generator that can be applied to any CNN directly during the training process without increasing the order of training time required. It is based on identifying a small number of the most uncertain base samples close to the estimated class boundaries and using them as seeds for augmentation. The resulting images are of high visual quality preserving small class differences, and they also improve the classifier boundary leading to higher recall scores than other state-of-the-art approaches.

AYURVEDIC MANAGEMENT OF POSTERIOR POLAR CATARACT: A CASE STUDY

Posterior Polar Cataract is the rare subset of congenital cataracts resulting in defective vision. It follows an autosomal dominant inheritance pattern and is occasionally sporadic due to the high incidence of intraoperative complications, which include posterior capsular rent, nucleus drop, etc. Posterior Polar Cataract is a surgical challenge for ophthalmologists. Ayurveda cataracts can be described in the context of Timira, Kacha and Linganasha. Timira is a disease of the Drishti Mandala, which presents with blurred vision initially, and if left untreated, it will progress to Kacha, which, if neglected, will lead to Linganasha. Linganasha is the last stage, where there is complete vision loss. A 37-year-old female patient complaining of blurred vision for distance and glare at night for two years is presented here. She was diagnosed with Posterior polar cataract one year back. The patient underwent Ayurveda treatment of shodhana and shamana chikitsa after properly assessing Rogi and Roga bala, which includes- Kayavirchana, Shirovirechana, Kriyakalpas (Akshitarpana, Anjana) and Rasayana prayogas. Assessment after the above treatment has shown improvement in the patient's visual acuity. From this study, it can be concluded that Ayurvedic treatment can preserve and give a better quality of vision to the patient of Posterior Polar Cataract, as surgery includes a high risk of capsular rupture and vitreous loss that can lead to worse visual outcomes.

High-resolution quantification of stress perfusion defects by cardiac magnetic resonance.

Quantitative stress perfusion cardiac magnetic resonance (CMR) is becoming more widely available, but it is still unclear how to integrate this information into clinical decision-making. Typically, pixel-wise perfusion maps are generated, but diagnostic and prognostic studies have summarized perfusion as just one value per patient or in 16 myocardial segments. In this study, the reporting of quantitative perfusion maps is extended from the standard 16 segments to a high-resolution bullseye. Cut-off thresholds are established for the high-resolution bullseye, and the identified perfusion defects are compared with visual assessment. Thirty-four patients with known or suspected coronary artery disease were retrospectively analysed. Visual perfusion defects were contoured on the CMR images and pixel-wise quantitative perfusion maps were generated. Cut-off values were established on the high-resolution bullseye consisting of 1800 points and compared with the per-segment, per-coronary, and per-patient resolution thresholds. Quantitative stress perfusion was significantly lower in visually abnormal pixels, 1.11 (0.75-1.57) vs. 2.35 (1.82-2.9) mL/min/g (Mann-Whitney U test P < 0.001), with an optimal cut-off of 1.72 mL/min/g. This was lower than the segment-wise optimal threshold of 1.92 mL/min/g. The Bland-Altman analysis showed that visual assessment underestimated large perfusion defects compared with the quantification with good agreement for smaller defect burdens. A Dice overlap of 0.68 (0.57-0.78) was found. This study introduces a high-resolution bullseye consisting of 1800 points, rather than 16, per patient for reporting quantitative stress perfusion, which may improve sensitivity. Using this representation, the threshold required to identify areas of reduced perfusion is lower than for segmental analysis.

Induced pluripotent stem cells as tool for cornea study in regenerative medicine

Induced Pluripotent Stem Cells technology as tool for cornea study in regenerative medicine.The ocular surface consists of an epithelial surface with specialized regions, including cornea lacrimal gland and conjunctiva. Altogether, these regions play a role in vision maintenance by transmitting light and, as first eye layer, ensure protection for the whole eye.These major roles are subject to defective developments or imbalanced maintenance, due to genetic variants or cellular degeneration, respectively, causing vision‐related problems. Intensive research is performed world‐wide to develop new therapies for such vision defects by using in vitro models, but lack of eye donations or low viability is an obstacle for such assays. At the same time, variability among donors on their genetic or epigenetic background is the main barrier for reproducibility, in addition to advanced age of donor, which is a factor limiting cell expansion or present a high number of senescent cells.A promising approach is to use stem cells to generate ocular surface cells or tissue, by mimicking the embryo development. It exists multiple sources of stem cells able to in vitro models, adult corneal cells isolated from biopsies have the advantage to resemble the organ source, but the collection has the disadvantage to be invasive. An alternative and highly valuable approach to generate organoid is from pluripotent stem cells, carrying the potential to differentiate into most cell types of the body and are subdivided into embryonic stem cells, derived from the inner cell mass of preimplantation embryos, and induced pluripotent stem cells (IPSCs) reprogrammed from adult cells.As IPSCs are one of the most promising approaches to counteract this lack of donations, several assays were performed to select the best conditions to generate cornea cells. The main goal is to characterize the developmental procedure by growth factors modulation and assess the maturation degree for the final product. The most important aspect to generate such final product is the heterogeneity of cell population during differentiation or maturation, as epithelial cornea layer has a different cell origin in comparison to stroma or endothelial cornea.Once differentiation and maturation are achieved, this in vitro model can be used for multiple downstream applications. As therapeutic approaches, organoids can be used as a living biobank or to model corneal diseases, either by using cells from affected patients or by base editing, introducing a causative mutation into the genome. Either patient‐derived or control organoids have the high potential to develop gene therapy, optogenetics, be a source of transplantable cells to replace generating tissue or screen small molecules treatments.For this topic, we will focus on the IPSC technology to derive cornea tissue and its perspective for eye development studies, clinical or preclinical study by comparing the methodology depending of the potential applications.

Biomarkers in retinal detachment surgery. How can we improve visual outcomes

Rhegmatogenous retinal detachment (RRD) is associated with separation of the neurosensory retina from the RPE, leading to visual field defect and potential reduction of central vision. Despite significant improvement in primary and final reattachment rate exceeding 90% and 98% respectively, functional outcomes are still difficult to predict and vary widely. In cases with macular involvement several factors have been implicated in the final visual recovery. These biomarkers are either related to individual risk factors (e.g., duration of symptoms, age, high myopia, and concomitant retinal disease),or to peri and postoperative factors (time to surgery, type of surgery, day of the week, experience of surgeon, positioning before and after surgery, and follow‐up management). With the availability of high‐resolution imaging modalities additional biomarkers have emerged, facilitating further more accurate prediction of visual outcome. Incorporating retinal detachment biomarkers into the daily clinical practice is essential to provide more effective patient counselling and potentially modify the visual rehabilitation process.

Visual Defect Detection and Analysis of Digital Robot Based on Virtual Artificial Intelligence Algorithm

The current research background of digital robot visual defect detection focuses on the application of virtual artificial intelligence algorithms. Convolutional neural networks (CNNS) perform well in the field of image processing and can learn and extract image features, which provides a more refined analysis ability for the visual defect detection of digital robots. Therefore, this paper focuses on the application of convolutional neural networks in digital robot vision defect detection, and explores the main processes of data collection and preprocessing, feature extraction, model training and defect detection in detail. Finally, the results of simulation experiments are as follows: Compared with traditional methods that rely on edge detection or threshold segmentation, the optimized defect detection accuracy rate is increased by 11.4%, and the image detection speed is increased by about 33.9%. The research results have important practical significance for improving the quality control of industrial manufacturing, which can not only improve the quality of products, but also adapt to the constant changes of the production line.

Research on defect recognition technology of transmission line based on visual macromodeling

Abstract In order to improve the defect recognition efficiency of transmission lines, the industry is currently using aerial images for automatic visual defect detection to ensure the safe operation of transmission lines. This paper proposes a method for defect recognition from coarse to fine, based on convolutional neural networks and connected domain algorithms, to improve recognition accuracy. The recognition speed is improved by using the knowledge distillation method of target detection networks based on decoupled features, adversarial features, and attention features. It has been found that the optimized recognition model improves the precision rate by 7%, the recall rate by 8%, and the average accuracy rate by 10%. The FPS of the model optimized by knowledge distillation is 62.5, and the average value of the FPS of other versions of this model is 47.35. It is believed that the two optimization ideas introduced in this paper can enhance the previous transmission line defect recognition algorithm in terms of accuracy and recognition speed.

Representation Learning Based on Co-Evolutionary Combined With Probability Distribution Optimization for Precise Defect Location.

Visual defect detection methods based on representation learning play an important role in industrial scenarios. Defect detection technology based on representation learning has made significant progress. However, existing defect detection methods still face three challenges: first, the extreme scarcity of industrial defect samples makes training difficult. Second, due to the characteristics of industrial defects, such as blur and background interference, it is challenging to obtain fuzzy defect separation edges and context information. Third, industrial defects cannot obtain accurate positioning information. This article proposes feature co-evolution interaction architecture (CIA) and glass container defect dataset to address the above challenges. Specifically, the contributions of this article are as follows: first, this article designs a glass container image acquisition system that combines RGB and polarization information to create a glass container defect dataset containing more than 60 000 samples to alleviate the sample scarcity problem in industrial scenarios. Subsequently, this article designs the CIA. CIA optimizes the probability distribution of features through the co-evolution of edge and context features, thereby improving detection accuracy in blurred defects and noisy environments. Finally, this article proposes a novel inforced IoU loss (IIoU loss), which can obtain more accurate position information by being aware of the scale changes of the predicted box. Defect detection experiments in three mainstream industrial manufacturing categories (Northeastern University (NEU)-Det, glass containers, wood) show that CIA only uses 22.5 GFLOPs, and mean average precision (mAP) (NEU-Det: 88.74%, glass containers: 95.38%, wood: 68.42%) outperforms state-of-the-art methods.

What OPA1 models teach us about potential therapy

Autosomal dominant optic atrophy (DOA) is the most common inherited optic neuropathy, affecting approximately 1 in 25 000 people. DOA has an insidious onset in early childhood. It presents typically with bilateral central painless vision loss, dyschromatopsia, and optic disc pallor due to gradual retinal ganglion cell (RGC) loss and optic nerve degeneration. 60%–70% of genetically confirmed DOA cases are associated with variants in OPA1. OPA1 is an ubiquitously expressed GTPase that regulates mitochondrial homeostasis through coordination of inner membrane fusion, maintenance of cristae structure, and maintenance of bioenergetic output. Over 250 known pathogenic variants in OPA1 have been reported, with missense variants within the GTPase domain associated with an increased susceptibility to developing DOA plus (DOA+), a multisystemic syndromic form of the disease with extraocular features, including sensorineural hearing loss and ataxia. Current therapeutic options to treat DOA are limited. Idebenone has been shown to protect against visual loss in Leber's hereditary optic neuropathy (LHON). In DOA off ‐label retrospective series show some promise in some patients.Animal models for OPA1 DOA, have been reported, including Drosophila, C elegans, Danio rerio and mouse models. They have cast light on some potential therapeutic modalities. Despite limitations, murine models of DOA are currently indispensable for understanding pathophysiology of disease, drug testing and development. They replicate many symptoms and signs of human disease. However, there are caveats. Chemically induced Complex I deficits are very toxic. Genetic models of mitochondrial mutation affect protein integrity and function and can cause severe pathologies. This is made more problematic if there is a need for aging the mouse to bring out the full phenotype. We shall review the mouse models and describe some features that aid or our understanding of pathophysiological mechanisms. Our research has characterized the earliest morphological changes as a progressive loss of synapses, dendrite retraction and RGC remodelling and loss of connectivity. In the DOA mouse Opa1+/− mouse (B6;C3‐Opa1Q285STOP), heterozygous Opa1+/− mice develop visual defects, on psychophysics (optokinetic response) and electrophysiology (VEP, PhNR), consistent with RGC dysfunction. There is depressed mitochondrial Complex I and IV activity. Visible structural changes and disrupted axon morphology in the optic nerve are seen. We have carried out a randomized, placebo‐controlled trial of the benzoquinone idebenone in our ADOA mouse, which revealed a limited therapeutic effect on RGC dendropathy and visual function.It is apparent that there are large differences between the rodent and primate visual system. The fovea of the retina, a critical component for central vision is one of the only primate‐specific components of the mammalian nervous system and is thus, not found in rodent. For these reasons, there is a strong case to use human cells. We shall review data on induced human pluripotent stem cell (hiPSC) with OPA1 mutation, with or without differentiation to RGCs, as models, and on CRISPRCas9 correction of patient‐derived OPA1 iPSC lines.

Adaptive explainable artificial intelligence for visual defect inspection.

Explainable Artificial Intelligence promises to deliver means so that humans better understand the rationale behind a particular machine learning model. In the image domain, such information is frequently conveyed through heat maps. Along the same line, information regarding defect detection for unsupervised methods applied to images can be conveyed through anomaly maps. Nevertheless, heat maps or anomaly maps can convey inaccurate information (artifacts), or their perceptions may differ across different persons. Therefore, the user experience could be enhanced by collecting human feedback and creating predictive models on how these could be recolored to bridge the gap between the original heat maps and anomaly maps created with explainability techniques and the output expected by humans. We envision this work as relevant in at least two scenarios. First, enhance anomaly and heat maps when conveying information regarding machine vision models deployed in production to remove information deemed unnecessary by the user but systematically present through the explainability technique due to underlying model issues (artifacts). Second, adapt anomaly and heat maps based on users’ perceptual needs and preferences.

RNA-sequencing expression profile and functional analysis of retinal pigment epithelium in atrophic age-related macular degeneration.

The retinal pigment epithelium (RPE) is fundamental to sustaining retinal homeostasis. RPE abnormality leads to visual defects and blindness, including age-related macular degeneration (AMD). Although breakthroughs have been made in the treatment of neovascular AMD, effective intervention for atrophic AMD is largely absent. The inadequate knowledge of RPE pathology is hindered by a lack of patient RPE datasets, especially at the single-cell resolution. In this study, we delved into a large-scale single-cell resource of AMD donors in which RPE cells were occupied in a substantial proportion. Bulk RNA-seq datasets of atrophic AMD were integrated to extract molecular characteristics of RPE in the pathogenesis of atrophic AMD. Both in vivo and in vitro models revealed that carboxypeptidase X, M14 family member 2 (CPXM2) was specifically expressed in the RPE cells of atrophic AMD, which might be induced by oxidative stress and involved in the epithelial-mesenchymal transition of RPE cells. Additionally, silencing of CPXM2 inhibited the mesenchymal phenotype of RPE cells in an oxidative stress cell model. Thus, our results demonstrate that CPXM2 plays a crucial role in regulating atrophic AMD and may serve as a potential therapeutic target for atrophic AMD.

Acute bilateral toxic endothelitis in binge alcohol consumption – A case series

Endothelitis has a vast range of etiologies, including infective, chemical-induced, drug-induced, systemic, and toxic causes. Toxic endothelitis secondary to alcohol binge consumption is rare and is reported less frequently in the literature. We report a series of three cases with bilateral defective vision due to corneal edema and endothelitis following binge alcohol consumption. All cases showed dramatic improvement with frequent topical steroid application alone. We hypothesize the cause for corneal edema as transient suppression of corneal endothelial cell metabolism rather than apoptosis, with later regain of functions following alcohol cessation and control of further inflammatory insults with intensive steroids.

Prevalence, clinical presentation, management of pituitary tumour and its complications among elderly population in Mumbai, India.

Pituitary tumour is not typically thought of as an elderly patient's condition. Hence, we examined all cases of confirmed or suspected pituitary tumour diagnosed in a tertiary hospitals at Mumbai, India during May 2015 and May 2023 among patients over the age of 70 to evaluate the prevalence, clinical presentation, management, complications in elderly patients with a pituitary tumour. After the age of 70 years, 16% people having pituitary tumour were observed. The volume of fossa was statistically greater in elderly patients. The duration of follow up was statistically longer in younger controls. The visual defects observed in elderly group were greater than young patients. Pituitary adenomas in old patients can be treated with trans-sphenoidal-adenomectomy. However, the proportion is lower than younger controls. Data shows that post-operative radiotherapy was more commonly observed in old patients with pituitary adenoma than younger controls.

Vision-based strawberry classification using generalized and robust deep networks

Grading of agricultural products such as fruits and vegetables based on ripeness level and visual defects for the purpose of export, storage and waste control is a process of special importance. Various methods have been used to detect levels of ripeness and the quality of agricultural products, some of which are destructive and some non-destructive. The machine vision system is one of the non-destructive and accurate systems in the field of detecting the quality of agricultural products. In this study, we propose a robust and generalized model via fine-tuning the pre-trained networks for the classification of strawberry fruit. A dataset containing 800 confirmed strawberry images in four classes (unripe, half-ripe, ripe, and damaged) was used. Instead of using fundamental data augmentation (FDA) techniques to prevent overfitting problem and increase the robustness of the model, we employed a novel learning-to-augment strategy (LAS) using noisy images that creates new noisy variant of data via original images. By using the Bayesian optimization algorithm, controllers were used to select the optimal noise parameters of Gaussian and speckle noise to generate new noise images. The best policies of data augmentation based on LAS was used to fine-tune pre-trained cutting-edge models (GoogleNet, ResNet18, and ShuffleNet). The results show that in all the proposed scenarios (i.e. using original data without data augmentation, employing FDA, and applying LAS) the GoogleNet model was able to achieve 96.88 %, 97.50 %, and 98.85 % accuracy, respectively.

PRESERFLO MicroShunt implantation versus trabeculectomy for primary open-angle glaucoma: a two-year follow-up study

BackgroundTo compare the intermediate-term efficacy of PRESERFLO (PF) MicroShunt implantation with trabeculectomy (TE) in patients with primary open-angle glaucoma, focusing on longitudinal changes of functional and structural parameters.MethodsThis retrospective comparative study included 104 eyes of 104 patients who underwent TE and 83 eyes of 83 patients that underwent PF implantation between January 2019 and December 2020, with a minimum follow-up of two years. Baseline and postoperative intraocular pressure (IOP), number of IOP-lowering medications, visual field mean defect (MD) and peripapillary retinal nerve fibre layer (RNFL) thickness measured using optical coherence tomography were assessed and compared between groups.ResultsBaseline characteristics (age, sex, IOP, number of IOP-lowering medications, MD, RNFL thickness) were comparable between the two groups (all P > 0.05). During the two-year of follow-up, mean IOP decreased from 24.09 ± 1.15 mmHg and 21.67 ± 0.77 mmHg to 11.37 ± 1.13 mmHg (P < 0.001) and 15.50 ± 1.54 mmHg (P = 0.028), and the mean number of IOP-lowering medications decreased from 3.25 ± 0.14 and 3.51 ± 0.14 to 0.53 ± 0.14 (P < 0.001) and 1.06 ± 0.43 (P < 0.001) in the TE and PF groups, respectively. MD remained stable [− 11.54 ± 0.93 dB and − 11.17 ± 1.66 to − 10.67 ± 0.91 dB (P = 0.226) and − 10.40 ± 4.75 dB (P = 0.628) in the TE and PF groups, respectively] but RNFL thickness decreased continuously during follow-up [62.79 ± 1.94 µm and 62.62 ± 2.05 µm to 57.41 ± 1.81 µm (P < 0.001) and 60.22 ± 1.98 µm (P = 0.182) in the TE and PF groups, respectively].ConclusionPF implantation is comparably effective in the intermediate term in lowering IOP and reducing the use of IOP-lowering medications over a two-year follow-up period. Although visual field defects were stable, RNFL continued to decrease during postoperative follow-up.

Autoencoder-Based Visual Anomaly Localization for Manufacturing Quality Control

Manufacturing industries require the efficient and voluminous production of high-quality finished goods. In the context of Industry 4.0, visual anomaly detection poses an optimistic solution for automatically controlled product quality with high precision. In general, automation based on computer vision is a promising solution to prevent bottlenecks at the product quality checkpoint. We considered recent advancements in machine learning to improve visual defect localization, but challenges persist in obtaining a balanced feature set and database of the wide variety of defects occurring in the production line. Hence, this paper proposes a defect localizing autoencoder with unsupervised class selection by clustering with k-means the features extracted from a pretrained VGG16 network. Moreover, the selected classes of defects are augmented with natural wild textures to simulate artificial defects. The study demonstrates the effectiveness of the defect localizing autoencoder with unsupervised class selection for improving defect detection in manufacturing industries. The proposed methodology shows promising results with precise and accurate localization of quality defects on melamine-faced boards for the furniture industry. Incorporating artificial defects into the training data shows significant potential for practical implementation in real-world quality control scenarios.

Surgical Outcome in Patients with Traumatic and Atraumatic Optic Neuropathy Following Endoscopic Optic Nerve Decompression.

Endoscopic optic nerve decompression is a highly effective and reliable approach for patients with select cases of optic neuropathy. It reduces the hydrostatic pressure and helps in relieving the compressive forces within the intracanalicular portion of the optic nerve consequently preserving and arresting the deterioration of vision. The advantages of the endoscopic approach over other procedures include preservation of olfaction, quick recovery, lack of external scars and less morbidity. The objective of the study isto assess the surgical outcome of endoscopic optic nerve decompression in patients with traumatic and atraumatic optic neuropathy. The case records of consequent patients attending ENT department with defective vision due to traumatic and atraumatic causes were reviewed. The outcome was measured in terms of improvement in visual acuity by log MAR scale. Among the 14 patients studied, 57% were females and the median age of presentation was 33 years. As compared to preoperative baseline visual acuity, the overall improvement was achieved in 11 patients, with an effective rate of 78.5% postoperatively. Patients presenting with no light perception and residual vision had significant improvement in visual acuity after surgery. Trans-nasal endoscopic surgery helps in decompressing the optic nerve with proper exposure of the orbital apex and optic canal without any intracranial or intra-orbital complications. Further, being a minimally invasive procedure has great advantage in cases of optic neuropathy.

An effective MID-based visual defect detection method for specular car body surface

The quality control of car body surfaces is one of the most important tasks in automotive manufacturing, which can directly affect the appearance of cars and purchasing experience of consumers. However, current car body defect detection methods rely mostly on manual inspection, which is subjective, inaccurate, and labor intensive. One of the biggest challenges of automated defect detection is that the specular reflection properties of car body surfaces make obtaining high-quality defect images extremely difficult. This article proposes a novel method called modulated intensity decoding (MID), in which preset encoded fringe patterns are projected onto the surface of a car body, and a camera is used to capture images after the patterns are reflected by the surface. By decoding the captured reflection images, high-quality surface defective decoded (SDD) images are obtained, which preserve the appearance of the defects and effectively highlight the defects against the background. Comparison experiments are performed on detecting and classifying the car body surface defects using four typical object detection networks to demonstrate the effectiveness of the proposed method. The results indicate that our method is far superior to that of conventional approaches using a square-wave-like-pattern light source.

Computer vision defect detection on unseen backgrounds for manufacturing inspection

Visual defect detection is an important aspect of quality inspection. High performance manufacturing can benefit from automating defect detection, and deep learning techniques are the current state of the art for this computer vision task. The premise of deep learning is that an over parameterized model learns to generalize in performing a task such as classification or object detection by exposure to a wide variety of training examples. In the canonical example, a classifier that has seen thousands of pictures of dogs and cats in different situations and backgrounds will be able to generalize to tell one animal from the other in a newly obtained photo. The assumption that the training data contains great variety is not met in typical manufacturing data sets. Data is highly repetitive and mostly represents defect-free products, meaning there are few images of defects or deviations to learn from. In this training regime, deep learning models are easily over-fit to the training data and can fail to detect defects in the face of variations such as position, lighting, or data drift. In this work, we explore training defect detection models with images of defects on different backgrounds and in different locations, in order to approximate the exposure to highly diverse data sets that is an assumption of a well-trained deep learning model. We demonstrate how models trained on diverse images containing a common defect type can pick defects out in new circumstances. Such generic models could be more robust to new defects not found in data collected for training, and can reduce data collection impediments to implementing visual inspection on production lines. Additionally, we demonstrate that object detection models trained to predict a label and bounding box outperform classifiers that predict a label only on held out test data typical of manufacturing inspection tasks. Finally, we studied the factors that affect generalization in order to train models that work under a wider range of conditions.