README.md

---
license: apache-2.0
task_categories:
- object-detection
tags:
- COCO
- Detection
- '2017'
pretty_name: COCO detection dataset script
size_categories:
- 100K<n<1M
dataset_info:
  config_name: '2017'
  features:
  - name: id
    dtype: int64
  - name: objects
    struct:
    - name: bbox_id
      sequence: int64
    - name: category_id
      sequence:
        class_label:
          names:
            '0': N/A
            '1': person
            '2': bicycle
            '3': car
            '4': motorcycle
            '5': airplane
            '6': bus
            '7': train
            '8': truck
            '9': boat
            '10': traffic light
            '11': fire hydrant
            '12': street sign
            '13': stop sign
            '14': parking meter
            '15': bench
            '16': bird
            '17': cat
            '18': dog
            '19': horse
            '20': sheep
            '21': cow
            '22': elephant
            '23': bear
            '24': zebra
            '25': giraffe
            '26': hat
            '27': backpack
            '28': umbrella
            '29': shoe
            '30': eye glasses
            '31': handbag
            '32': tie
            '33': suitcase
            '34': frisbee
            '35': skis
            '36': snowboard
            '37': sports ball
            '38': kite
            '39': baseball bat
            '40': baseball glove
            '41': skateboard
            '42': surfboard
            '43': tennis racket
            '44': bottle
            '45': plate
            '46': wine glass
            '47': cup
            '48': fork
            '49': knife
            '50': spoon
            '51': bowl
            '52': banana
            '53': apple
            '54': sandwich
            '55': orange
            '56': broccoli
            '57': carrot
            '58': hot dog
            '59': pizza
            '60': donut
            '61': cake
            '62': chair
            '63': couch
            '64': potted plant
            '65': bed
            '66': mirror
            '67': dining table
            '68': window
            '69': desk
            '70': toilet
            '71': door
            '72': tv
            '73': laptop
            '74': mouse
            '75': remote
            '76': keyboard
            '77': cell phone
            '78': microwave
            '79': oven
            '80': toaster
            '81': sink
            '82': refrigerator
            '83': blender
            '84': book
            '85': clock
            '86': vase
            '87': scissors
            '88': teddy bear
            '89': hair drier
            '90': toothbrush
    - name: bbox
      sequence:
        sequence: float64
        length: 4
    - name: iscrowd
      sequence: int64
    - name: area
      sequence: float64
  - name: height
    dtype: int64
  - name: width
    dtype: int64
  - name: file_name
    dtype: string
  - name: coco_url
    dtype: string
  - name: image_path
    dtype: string
  splits:
  - name: train
    num_bytes: 87231216
    num_examples: 117266
  - name: validation
    num_bytes: 3692192
    num_examples: 4952
  download_size: 20405354669
  dataset_size: 90923408
---
## Usage
For using the COCO dataset (2017), you need to download it manually first:
```bash
wget http://images.cocodataset.org/zips/train2017.zip
wget http://images.cocodataset.org/zips/val2017.zip
wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip
```

Then to load the dataset:
```python
COCO_DIR = ...(path to the downloaded dataset directory)...
ds = datasets.load_dataset(
    "yonigozlan/coco_2017_detection_script",
    "2017",
    data_dir=COCO_DIR,
    trust_remote_code=True,
)
```

## Benchmarking
Here is an example of how to benchmark a 🤗 Transformers object detection model on the validation data of the COCO dataset:

```python
import datasets
import torch
from PIL import Image
from torch.utils.data import DataLoader
from torchmetrics.detection.mean_ap import MeanAveragePrecision
from tqdm import tqdm

from transformers import AutoImageProcessor, AutoModelForObjectDetection

# prepare data
COCO_DIR = ...(path to the downloaded dataset directory)...
ds = datasets.load_dataset(
    "yonigozlan/coco_2017_detection_script",
    "2017",
    data_dir=COCO_DIR,
    trust_remote_code=True,
)
val_data = ds["validation"]
categories = val_data.features["objects"]["category_id"].feature.names
id2label = {index: x for index, x in enumerate(categories, start=0)}
label2id = {v: k for k, v in id2label.items()}
checkpoint = "facebook/detr-resnet-50"

# load model and processor
model = AutoModelForObjectDetection.from_pretrained(
    checkpoint, torch_dtype=torch.float16
).to("cuda")
id2label_model = model.config.id2label
processor = AutoImageProcessor.from_pretrained(checkpoint)


def collate_fn(batch):
    data = {}
    images = [Image.open(x["image_path"]).convert("RGB") for x in batch]
    data["images"] = images
    annotations = []
    for x in batch:
        boxes = x["objects"]["bbox"]
        # convert to xyxy format
        boxes = [[box[0], box[1], box[0] + box[2], box[1] + box[3]] for box in boxes]
        labels = x["objects"]["category_id"]
        boxes = torch.tensor(boxes)
        labels = torch.tensor(labels)
        annotations.append({"boxes": boxes, "labels": labels})
    data["original_size"] = [(x["height"], x["width"]) for x in batch]
    data["annotations"] = annotations
    return data


# prepare dataloader
dataloader = DataLoader(val_data, batch_size=8, collate_fn=collate_fn)

# prepare metric
metric = MeanAveragePrecision(box_format="xyxy", class_metrics=True)

# evaluation loop
for i, batch in tqdm(enumerate(dataloader), total=len(dataloader)):
    inputs = (
        processor(batch["images"], return_tensors="pt").to("cuda").to(torch.float16)
    )
    with torch.no_grad():
        outputs = model(**inputs)
    target_sizes = torch.tensor([x for x in batch["original_size"]]).to("cuda")
    results = processor.post_process_object_detection(
        outputs, threshold=0.0, target_sizes=target_sizes
    )

    # convert predicted label id to dataset label id
    if len(id2label_model) != len(id2label):
        for result in results:
            result["labels"] = torch.tensor(
                [label2id.get(id2label_model[x.item()], 0) for x in result["labels"]]
            )
    # put results back to cpu
    for result in results:
        for k, v in result.items():
            if isinstance(v, torch.Tensor):
                result[k] = v.to("cpu")
    metric.update(results, batch["annotations"])

metrics = metric.compute()
print(metrics)
```