The authors evaluated intermediate activations from DiT and SiT-style diffusion backbones under the same downstream probing protocol, producing a performance profile across layers.
Diffusion transformer depth profiles reflect prediction targets, not just feature granularity
Layer-by-layer probes suggest that semantic representations emerge temporarily as diffusion transformers construct their trained prediction targets.
Chinese Tech
Benyuan Meng · Longtao Huang · Yang Liu · Liang Lin
Sun Yat-sen University · Alibaba Group · Peng Cheng Laboratory · X-Era AI Lab
Research Digest··3 min read
Meng and colleagues re-examine why intermediate features in diffusion transformers improve and then deteriorate on downstream vision benchmarks as network depth increases.
Why this paper
From Alibaba Group and 3 others
In one line
The inverted-U depth profile in diffusion transformers reflects where the prediction target is encoded, not a coarse-to-fine semantic signal.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§