April 23, 2025
April 23, 2025

Snap Ads Understanding

From Pixels to Words

References

1

Radford, Alec, et al. "Learning transferable visual models from natural language supervision." International conference on machine learning. PmLR, 2021.

2

The embedding is computed as the mean-pool of 16 frames, sampled at one frame per second over the first 16 seconds of the CLIP embedding. We observed similar results using other pooling mechanisms.

References
1

Radford, Alec, et al. "Learning transferable visual models from natural language supervision." International conference on machine learning. PmLR, 2021.

2

The embedding is computed as the mean-pool of 16 frames, sampled at one frame per second over the first 16 seconds of the CLIP embedding. We observed similar results using other pooling mechanisms.