model card

DIAG-1

resnet18 · thfood100 · 100 classes · submitted 2026-08-05 04:42:16

Rank · Score
#2 · 90.0

Training configuration

Architecture
resnet18
Dataset
thfood100
Epochs trained
5
Optimizer
AdamW
Learning rate
0.0003
Batch size
64
Scheduler
CosineAnnealingLR
Mixed precision
on
Seed
42

Measured on this server

Validation accuracy
88.56%
Parameters
11,227,812
Model size
42.9 MB
Latency (batch 1)
48.9 ms
Throughput
20.4 FPS
Inference memory (approx.)
42.9 MB
Try this model Leaderboard