computer vision product teams

Abstract: We present Scalable Pixel-anchored End-to-end Diffusion (SpeeDiff), a latent diffusion method that jointly trains the VAE and the diffusion model from scratch. In principle, joint training allows the diffusion loss gradient to directly guide the VAE encoder, encouraging the formation of a generation-friendly latent space and potentially yielding faster convergence than the conventional two-stage approach with a pretrained frozen VAE. However, a naive end-to-end implementation severely degrades performance, as unrestricted backpropagation of the diffusion loss leads to latent space collapse. Our main technical contribution is a simple yet effective Tweedie Pixel Reconstruction (TPR) loss, which provides additional pixel-level feedback by decoding a predicted clean latent from an intermediate noisy state using Tweedie's formula, thereby alleviating collapse. Furthermore, our method enables jointly scaling a fully transformer-based architecture and enhances representation alignment within the end-to-end framework. Our SpeeDiff-XL model achieves over 140x and 61x faster training compared to Vanilla SiT and REPA, respectively, while attaining an FID of 1.50 without guidance on ImageNet 256x256 generation. With a more efficient 32x compressed VAE, our model further reaches an FID of 1.53 without guidance on ImageNet 512x512 generation.

Abstract: Despite the success of foundation models in image synthesis, achieving fine-grained spatial control remains a significant challenge, typically requiring auxiliary structural adapters or complex inference-time steering. While modern vision-language (VLM) text encoders possess the latent capacity to represent spatial coordinates, generative models often fail to leverage this information. We demonstrate that the signal-to-noise ratio (SNR) distribution is a critical determinant of spatial grounding in flow-matching models. Using a novel synthetic benchmark of geometric primitives, we isolate structural mechanics from textural complexity inherent in natural imagery and introduce a rigorous color-aware layout metric to quantify spatial fidelity. We show that training on a centered SNR distribution results in poor layout adherence (25.9%). Instead, we identify high-noise regimes as a "critical learning phase" for spatial grounding. Shifting the training focus toward these stages increases layout adherence to 80.8%, effectively reaching the 83.6% reconstruction ceiling of the underlying variational autoencoder. We find that the necessity for this shift scales with resolution and aspect ratio variance, and that the conservative noise schedules employed by models like FLUX.1 are insufficient for precise layout adherence. Finally, we show that these findings translate to real-world imagery via the COCO dataset (1.0 vs. 57.3 Average Precision), where state-of-the-art layout adherence is unlocked through distribution tuning alone, without architectural modifications.

Abstract: Multimodal Large Language Models (MLLMs) encode images into dense visual token sequences, creating substantial attention and KV-cache overhead. Existing visual token compression methods reduce this cost by pruning low-saliency tokens in shallow layers, but this practice couples temporary computational deactivation with irreversible information deletion. We show that this coupling breaks an intrinsic inference pattern in MLLMs. Through layer-wise attention analysis, we identify \textbf{Visual Backtracking}: visual pivots remain spatially stable, yet in middle layers they actively retrieve context from tokens that appeared unimportant earlier. Thus, early low-saliency tokens can be inactive for computation while still being useful for later reasoning. Motivated by this finding, we propose \textbf{ReTrace}, a backtracking-aware compression framework that introduces a dormant token state. ReTrace caches non-pivot tokens outside expensive layer-wise computation, restores them during the backtracking window, and prunes only after their information has been absorbed. Experiments on Qwen2.5-VL and InternVL3 demonstrate nearly 90\% reductions in FLOPs and KV cache, a 2.83$\times$ prefilling speedup, and strong performance retention across complex vision-language benchmarks.

Abstract: When a vision-language model is finetuned on class-imbalanced data, rare-class recall can fall well below the model's own zero-shot performance. In this regime, we uncover a surprising matched-prompt penalty whereby the standard choice of reusing the descriptive finetuning prompt at inference can yield lower rare-class recall than using a different descriptive prompt. We call this inference-time strategy cross-prompt inference. We evaluate diverse VLMs across six datasets and find that cross-prompt inference consistently recovers rare-class recall, with varying trade-offs in frequent-class recall. The effect reflects a train-eval prompt interaction rather than prompt quality alone, and does not arise with non-descriptive prompt changes. It is strongest when imbalance and visual confusability produce substantial rare-class suppression, and is weak or absent when such suppression does not occur. Using Tuning Contribution (TuCo), which decomposes a model's computation into pretrained and finetuning parts, we find that cross-prompt inference reduces reliance on finetuning-induced computation and shifts the model toward its pretrained component. Reducing that finetuning contribution also substantially reduces the cross-prompt advantage. These results show that rare-class knowledge suppressed by imbalanced finetuning can remain accessible under a different inference prompt.