Webcam eye tracking can greatly accelerate data collection in basic and applied research by reaching large, diverse samples of participants online. However, its spatial performance is considered well below that of hardware-based systems, limiting its use. We evaluated the first-generation Chiasm webcam eye tracker in a large online study (N=505) spanning diverse participants, devices, and recording conditions. Chiasm achieved a mean spatial accuracy of 2.39 cm (≈2.28∘ at 60 cm), spatial precision (SD) of 1.05 cm (≈1.00∘ at 60 cm), and sample-to-sample precision (RMS-S2S) of 1.29 cm (≈1.23∘ at 60 cm). For the 91% of participants who met a 3 cm calibration criterion, spatial accuracy was 2.15 cm (≈2.05∘ at 60 cm). After three minutes of a task that encouraged head movement, spatial accuracy declined by 0.59 cm, while the gaze center at each target shifted by a mean distance of only 0.37 cm. On the validation block, the offset was 0.30 cm larger for men than for women. Age, region of residence, face luminance, and Michelson contrast showed no reliable relationship with spatial accuracy. A higher eye aspect ratio, a more open eye, was associated with a smaller offset. Wearing glasses decreased spatial accuracy on every validation block. In a comparison between a random subsample (N=101) and a separate sample tested with the open-source tracker WebGazer (N=101) using an identical protocol, Chiasm showed better spatial accuracy (2.18 vs. 4.51 cm) and spatial precision (SD; 1.01 vs. 1.55 cm). WebGazer showed better sample-to-sample precision (RMS-S2S; 0.79 vs. 1.26 cm), which likely reflects conservative, over-smoothed estimates rather than superior stability. These results suggest that Chiasm is narrowing the gap between webcam and hardware-based eye tracking, making large-scale gaze research "in the wild" a reality.
Figures & tables
Figure 1: The block order in both the Chiasm and WebGazer experiments. Every participant followed this sequence. Validation B repeated Validation A after the shooting game.
Accuracy
Precision (SD)
RMS-S2S
Calibration
Block
n
Mean (cm)
n
Mean (cm)
n
Mean (cm)
All
validation
502
2.39
502
1.05
502
1.29
Validation A
502
3.18
501
1.08
501
1.34
Validation B
499
3.77
496
1.11
496
1.36
Passed
validation
460
2.15
460
1.02
460
1.26
Validation A
457
2.96
456
1.06
456
1.32
Table 1: Mean evaluation metrics for Chiasm. Spatial accuracy, spatial precision (SD), and sample-to-sample precision (RMS-S2S) for all participants and those who passed the calibration threshold. n is the number of participants. n is smaller in some cells because a participant is omitted when fewer than six points on the validation block, or fewer than nine on Validation A or Validation B, yield a finite value for that metric.
Figure 2: Chiasm spatial accuracy at each true target, with the predicted center marked.
Figure 3: Empirical distribution of Chiasm spatial accuracy.
Figure 4: Chiasm within-person spatial precision (SD) drawn at each true target.
Figure 5: Chiasm drift on the dot grid. Each arrow runs from the validation A predicted center to the validation B center. The arrows are drawn at seven times their length for better readability.
Figure 6: Spatial accuracy distributions. Solid lines are Chiasm and dashed lines are WebGazer. Colour is the validation phase.
Accuracy
Precision (SD)
RMS-S2S
Study
Block
n
Mean (cm)
n
Mean (cm)
n
Mean (cm)
Chiasm
validation
101
2.18
101
1.01
101
1.26
Validation A
101
3.01
101
1.05
101
1.33
Validation B
100
3.62
100
1.09
100
1.36
WebGazer
validation
101
4.51
96
1.55
96
0.79
Validation A
98
5.62
94
1.61
94
0.91
Table 2: Mean evaluation metrics for the Chiasm subsample and the WebGazer study. Spatial accuracy, spatial precision (SD), and sample-to-sample precision (RMS-S2S). The Chiasm rows are the random subsample matched in size to the WebGazer study. n is the number of participants. n is smaller in some cells because a participant is omitted when fewer than six points on the validation block, or fewer than nine on Validation A or Validation B, yield a finite value for that metric.
Figure 7: Person-level means for the Chiasm subsample and the WebGazer study. (a) Spatial accuracy. (b) Spatial precision (SD). (c) Sample-to-sample precision (RMS-S2S). Each person is a point. The marker is the mean with its 95% confidence interval, and the bracket is an independent-samples t -test. Counts above 15 cm stay in the mean and the test.
Appendix figures & tables14 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 8: Attrition after the start of Validation A in the Chiasm study. Each bar counts participants who started that stage. Included are the participants who contributed to all validation blocks.
Figure 9: Attrition after the start of Validation A in the WebGazer study. Each bar counts participants who started that stage. Included are the participants who contributed to all validation blocks.
Figure 10: Computer, camera, and display reported by the 505 included participants.
Figure 11: Region of residence, gender, and age in the Chiasm included sample. A missing Prolific field is left off that panel.
Figure 12: Chiasm validation spatial accuracy by gender. A person with fewer than six usable dots is omitted.
Figure 13: Chiasm validation spatial accuracy by region of residence. A person with fewer than six usable dots is omitted.
Figure 14: Chiasm validation spatial accuracy by age. A person with fewer than six usable dots is omitted.
Figure 15: Chiasm mean spatial accuracy by glasses. Glasses is the session median of the face-frame glasses score, at least one half.
Figure 16: Chiasm mean spatial accuracy against eye aspect ratio on the same validation phase.
Figure 17: Chiasm mean spatial accuracy against mean face luminance on the same validation phase.
Figure 18: Chiasm mean spatial accuracy against Michelson contrast between the face and the surrounding frame.
Figure 19: Spatial accuracy at each true target for Chiasm and WebGazer.
Figure 20: Within-person spatial precision (SD) at each true target for Chiasm and WebGazer.
Figure 21: Drift on the dot grid, from the validation A predicted center to the validation B center for Chiasm and WebGazer.
This engineering study assesses gaze-assisted data acquisition during visual screening with a Gaze Quest implementation via webcam and GC308 near infrared camera with an Orlosky eye tracking pipeline. Ten subjects performed under both acquisition conditions. Accuracy of Tumbling E orientation, reported application logMAR value,and response latency were determined from keyboard responses and hence are behavioral outputs as opposed toeye tracker screening results. Frame-to-frame change in theangle between successive normalized gaze vectors (28.2degrees) served as the GC308 eye tracker derived diagnostic with an effective sampling rate of roughly 8 FPS. No screen target calibration or gaze to screen translation was done on the GC308 eye tracker output data for the current analysis and hence, these data represent acquisition stability and not gaze accuracy in relation to the target. The webcam pipeline relied on 25-point screen calibration. Because the display resolution and pixel pitch could not be retained in the records of the study, the reported application logMAR values cannot be viewed as measures of visual acuity in any way.
Geometric eye trackers can provide the spatial accuracy required for gaze-based interaction and multimodal studies, but their measurements remain sensitive to residual session-specific calibration error. Research on correcting this error is difficult to compare because methods are typically evaluated with different devices, target layouts, and error definitions. We present a calibration-focused dataset containing 163 trials from 12 participants, with separate 18-point fitting and 32-point test grids, and use it to evaluate global, local, and composite correction functions under a common spatial-extrapolation protocol. We further introduce a lightweight neural refiner that combines ranked predictions from complementary calibrators. On this controlled dataset, post-vendor correction reduces the mean angular error from 1.53∘ to 1.03∘ with the strongest classical composite and to 0.96∘ with the refiner. In a closed-loop gaze task, lower residual error is associated with higher performance across four online correction conditions. These results provide a reproducible data-quality benchmark for using gaze as a behavioral signal in interactive modeling.
Jiaqi Liu, Zixuan Wang, Yuhong Zhang +4
Institute for Neural Computation, University of California San Diego
Practical webcam gaze tracking is constrained not only by error, but also by calibration burden, robustness to head motion and session drift, runtime footprint, and browser use. We therefore target a deployment-oriented operating point rather than the image large-backbone regime. We cast landmark-based point-of-regard estimation as session-wise adaptation: a shared geometric encoder produces embeddings that can be aligned to a new session from a small calibration set. We present Equivariant Meta-Calibrated Gaze (EMC-Gaze), a lightweight landmark-only method combining an E(3)-equivariant landmark-graph encoder, local eye geometry, binocular emphasis, auxiliary 3D gaze-direction supervision, and a closed-form ridge calibrator differentiated through episodic meta-training. To reduce pose leakage, we use a two-view canonicalization consistency loss. The deployed predictor uses only facial landmarks and fits a per-session ridge head from brief calibration. In a fixation-style interactive evaluation over 33 sessions at 100 cm, EMC-Gaze achieves 5.79 +/- 1.81 deg RMSE after 9-point calibration versus 6.68 +/- 2.34 deg for Elastic Net; the gain is larger on still-head queries (2.92 +/- 0.75 deg vs. 4.45 +/- 0.30 deg). Across three subject holdouts of 10 subjects each, EMC-Gaze retains an advantage (5.66 +/- 0.19 deg vs. 6.49 +/- 0.33 deg). On MPIIFaceGaze with short per-session calibration, the eye-focused model reaches 8.82 +/- 1.21 deg at 16-shot calibration, ties Elastic Net at 1-shot, and outperforms it from 3-shot onward. The exported eye-focused encoder has 944,423 parameters, is 4.76 MB in ONNX, and supports calibrated browser prediction in 12.58/12.58/12.90 ms per sample (mean/median/p90) in Chromium 145 with ONNX Runtime Web. These results position EMC-Gaze as a calibration-friendly operating point rather than a universal state-of-the-art claim against heavier appearance-based systems.
Chenkai Zhang
Independent Researcher, Wenzhou, Zhejiang 325000, China