ACR Instance Segmentation RF-DETR v1.0.0

ACR ํ™˜๊ฒฝ์˜ ๊ฐ์ฒด ๊ฒ€์ถœ ๋ฐ instance segmentation์„ ์œ„ํ•œ RF-DETR Seg2XLarge ๋ฐฐํฌ ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค.

ํŒŒ์ผ

ํŒŒ์ผ ์„ค๋ช…
cmes_RF_960_v1.0.0.pth RF-DETR Seg2XLarge PyTorch best weight
config.yaml ๋ชจ๋ธ ๊ตฌ์กฐ, ์ถ”๋ก  ์„ค์ •, ํด๋ž˜์Šค ์ˆœ์„œ ๋ฐ ๊ฒ€์ฆ ๊ฒฐ๊ณผ

์ฃผ์š” ์„ค์ •

  • ์ž…๋ ฅ ์ด๋ฏธ์ง€: RGB
  • ์ž…๋ ฅ ํ•ด์ƒ๋„: 960 x 960
  • ๊ธฐ๋ณธ confidence threshold: 0.50
  • ์ตœ๋Œ€ detection ์ˆ˜: 300
  • ํด๋ž˜์Šค ์ˆ˜: 11
  • RF-DETR: 1.9.0

์„ค์น˜

python -m pip install rfdetr==1.9.0 torch torchvision pillow numpy pyyaml

๋‹ค์šด๋กœ๋“œ

from huggingface_hub import hf_hub_download

weight_path = hf_hub_download(
    repo_id="cmes-deepvision/ACR-instance-segmentation-RF-v1.0.0",
    filename="cmes_RF_960_v1.0.0.pth",
)

PyTorch ์ถ”๋ก 

import torch
from PIL import Image
from rfdetr import RFDETRSeg2XLarge

model = RFDETRSeg2XLarge.from_checkpoint(
    weight_path,
    trust_checkpoint=True,
    resolution=960,
    device="cuda:0",
)
model.inference(compile=False, batch_size=1, dtype=torch.float16)

image = Image.open("image.jpg").convert("RGB")
detections = model.predict(
    image,
    threshold=0.50,
    shape=(960, 960),
    patch_size=12,
    include_source_image=False,
)

์ด ํŒŒ์ผ์€ PyTorch pickle ๊ธฐ๋ฐ˜ ํ˜•์‹์ด๋ฏ€๋กœ ๊ณต์‹ ์ €์žฅ์†Œ์—์„œ ๋ฐ›์€ ํŒŒ์ผ์—๋งŒ trust_checkpoint=True๋ฅผ ์‚ฌ์šฉํ•˜์‹ญ์‹œ์˜ค. ํด๋ž˜์Šค ID๋Š” ์•„๋ž˜ ์ˆœ์„œ๋ฅผ ๋”ฐ๋ฆ…๋‹ˆ๋‹ค.

0  dropping item
1  dumping item
2  item in pb bag
3  item in tote
4  item on buffer
5  item on floor
6  item on plate
7  item out of buffer
8  item out of tote
9  picked item
10 tote

๊ฒ€์ฆ ๊ฒฐ๊ณผ

Test_v2.2.1 ์ „์ฒด 6,475์žฅ, 50 tasks, confidence 0.50, IoU 0.50 ๊ธฐ์ค€์ž…๋‹ˆ๋‹ค.

Precision Recall F1 bbox mAP50 segm mAP50
0.9054 0.8509 0.8657 0.8411 0.8454

์šด์˜ threshold๋ฅผ ๋ณ€๊ฒฝํ•˜๋ฉด Precision๊ณผ Recall๋„ ๋‹ฌ๋ผ์ง€๋ฏ€๋กœ ๊ธฐ๋ณธ๊ฐ’ 0.50์—์„œ ๋จผ์ € ๊ฒ€์ฆํ•˜์‹ญ์‹œ์˜ค.

Downloads last month
5
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Collection including cmes-deepvision/ACR-instance-segmentation-RF-v1.0.0