yolov7-1b1h-500ep – AI Model by dictvv | AlphaNeural AI

You can deploy this model and start earning money today!

dictvv/
yolov7-1b1h-500ep

like

0

object-detection

Views

No views yet

YOLOv7 1B1H 500 Epochs Training Results

Model Description

YOLOv7-tiny model trained on COCO 320x320 dataset for 500 epochs.

Training Configuration

Architecture: YOLOv7-tiny (1 Backbone + 1 Head)
Dataset: COCO 2017 (320x320 resized)
Epochs: 500
Batch Size: 128
Image Size: 320x320
Loss: non-OTA (ComputeLoss)
Optimizer: SGD
Learning Rate: 0.01

Results

Metric	Value
mAP@0.5	0.4365
mAP@0.6	0.3804
mAP@0.7	0.3135
mAP@0.8	0.2257
mAP@0.9	0.0975
mAP@0.5:0.95	0.2672
Precision	0.6479
Recall	0.3919

Files

weights/best.pt: Best checkpoint (stripped optimizer)
weights/last.pt: Final checkpoint (stripped optimizer)
weights/epoch_XXX.pt: Checkpoints with per-class AP data
results.txt: Training log with all metrics
hyp.yaml: Hyperparameters
opt.yaml: Training options

Usage

python

import torch
from models.experimental import attempt_load

# Load model
model = attempt_load('weights/best.pt', map_location='cuda')

# Inference
img = torch.randn(1, 3, 320, 320).cuda()
pred = model(img)

Per-class AP Access

python

import torch

# Load checkpoint with per-class AP
ckpt = torch.load('weights/epoch_499.pt', weights_only=False)

# Access per-class AP
per_class_ap = ckpt.get('per_class_ap', {})
for class_id, ap_data in per_class_ap.items():
    print(f"{ap_data['name']}: mAP@.5={ap_data['ap50']:.3f}, mAP@.7={ap_data['ap70']:.3f}")

Training Date

2025-12-22