# ------------------------------------------------------------------------ # RF-DETR # Copyright (c) 2025 Roboflow. All Rights Reserved. # Licensed under the Apache License, Version 2.0 [see LICENSE for details] # ------------------------------------------------------------------------ # Copied and modified from LW-DETR (https://github.com/Atten4Vis/LW-DETR) # Copyright (c) 2024 Baidu. All Rights Reserved. # ------------------------------------------------------------------------ """CLI orchestrator for ONNX and TensorRT model export.""" import os import random import numpy as np import torch import torch.nn as nn from PIL import Image from torchvision.transforms.v2 import Compose, Resize, ToDtype, ToImage from rfdetr.datasets.transforms import Normalize from rfdetr.export._onnx.exporter import export_onnx from rfdetr.export._tensorrt import trtexec from rfdetr.models import build_model from rfdetr.utilities.distributed import get_rank from rfdetr.utilities.logger import get_logger from rfdetr.utilities.package import get_sha, get_version logger = get_logger() def make_infer_image(infer_dir, shape, batch_size, device="cuda", num_channels: int = 3): if infer_dir is None: if num_channels == 3: dummy = np.random.randint(0, 256, (shape[0], shape[1], 3), dtype=np.uint8) image = Image.fromarray(dummy, mode="RGB") else: # Non-RGB: build a random float tensor directly, bypassing PIL. # Normalization is intentionally skipped here — export tracing only # requires tensors of the correct shape and dtype (float32), not the # correct distribution. Real inference normalizes via predict() before # the tensor reaches the exported model, so the ONNX/TensorRT graph # never sees raw [0, 1] inputs in production. inps = torch.rand(batch_size, num_channels, shape[0], shape[1], device=device) return inps else: if num_channels != 3: raise ValueError( "Providing `infer_dir` is only supported for RGB models (num_channels=3). " "For non-RGB models, omit `infer_dir` to use a synthetic dummy input." ) image = Image.open(infer_dir).convert("RGB") transforms = Compose( [ Resize((shape[0], shape[1])), ToImage(), ToDtype(torch.float32, scale=True), Normalize(), ] ) inps, _ = transforms(image, None) inps = inps.to(device) # inps = utils.nested_tensor_from_tensor_list([inps for _ in range(args.batch_size)]) inps = torch.stack([inps for _ in range(batch_size)]) return inps def no_batch_norm(model): for module in model.modules(): if isinstance(module, nn.BatchNorm2d): raise ValueError("BatchNorm2d found in the model. Please remove it.") def main(args): git_info = get_sha() if git_info != "unknown": logger.info(f"Running from git repository: {git_info}") else: version = get_version() logger.info(f"Running RF-DETR version: {version or 'unknown'}") logger.info(f"Export config: {vars(args)}") # convert device to device_id if args.device == "cuda": device_id = "0" elif args.device == "cpu": device_id = "" else: device_id = str(int(args.device)) args.device = f"cuda:{device_id}" # device for export onnx # TODO: export onnx with cuda failed with onnx error device = torch.device("cpu") os.environ["CUDA_VISIBLE_DEVICES"] = device_id # fix the seed for reproducibility seed = args.seed + get_rank() torch.manual_seed(seed) np.random.seed(seed) random.seed(seed) result = build_model(args) model = result[0] if isinstance(result, tuple) else result n_parameters = sum(p.numel() for p in model.parameters()) logger.info(f"number of parameters: {n_parameters}") n_backbone_parameters = sum(p.numel() for p in model.backbone.parameters()) logger.info(f"number of backbone parameters: {n_backbone_parameters}") n_projector_parameters = sum(p.numel() for p in model.backbone[0].projector.parameters()) logger.info(f"number of projector parameters: {n_projector_parameters}") n_backbone_encoder_parameters = sum(p.numel() for p in model.backbone[0].encoder.parameters()) logger.info(f"number of backbone encoder parameters: {n_backbone_encoder_parameters}") n_transformer_parameters = sum(p.numel() for p in model.transformer.parameters()) logger.info(f"number of transformer parameters: {n_transformer_parameters}") if args.resume: # weights_only=False: legacy checkpoints may contain non-tensor objects (argparse.Namespace). checkpoint = torch.load(args.resume, map_location="cpu", weights_only=False) result = model.load_state_dict(checkpoint["model"], strict=False) if result.missing_keys or result.unexpected_keys: logger.warning( "load_state_dict strict=False: missing=%s unexpected=%s", result.missing_keys[:5], result.unexpected_keys[:5], ) logger.info(f"load checkpoints {args.resume}") if args.layer_norm: no_batch_norm(model) model.to(device) input_tensors = make_infer_image(args.infer_dir, args.shape, args.batch_size, device) input_names = ["input"] if args.backbone_only: output_names = ["features"] elif args.segmentation_head: output_names = ["dets", "labels", "masks"] else: output_names = ["dets", "labels"] if getattr(args, "dynamic_batch", False): dynamic_axes = {name: {0: "batch"} for name in input_names + output_names} else: dynamic_axes = None # Run model inference in pytorch mode model.eval().to("cuda") input_tensors = input_tensors.to("cuda") with torch.no_grad(): if args.backbone_only: features = model(input_tensors) logger.debug(f"PyTorch inference output shape: {features.shape}") elif args.segmentation_head: outputs = model(input_tensors) dets = outputs["pred_boxes"] labels = outputs["pred_logits"] masks = outputs["pred_masks"] if isinstance(masks, torch.Tensor): logger.debug( f"PyTorch inference output shapes - Boxes: {dets.shape}, Labels: {labels.shape}, " f"Masks: {masks.shape}" ) else: # masks is a dict with spatial_features, query_features, bias logger.debug(f"PyTorch inference output shapes - Boxes: {dets.shape}, Labels: {labels.shape}") logger.debug( "Mask spatial_features: " f"{masks['spatial_features'].shape}, " f"query_features: {masks['query_features'].shape}, " f"bias: {masks['bias'].shape}" ) else: outputs = model(input_tensors) dets = outputs["pred_boxes"] labels = outputs["pred_logits"] logger.debug(f"PyTorch inference output shapes - Boxes: {dets.shape}, Labels: {labels.shape}") model.cpu() input_tensors = input_tensors.cpu() output_file = export_onnx( args.output_dir, model, input_names, input_tensors, output_names, dynamic_axes, backbone_only=args.backbone_only, verbose=args.verbose, opset_version=args.opset_version, variant_name=getattr(args, "variant_name", None), ) onnx_path = output_file # preserve ONNX path before any post-processing step overwrites it if args.tensorrt: output_file = trtexec(onnx_path, args) # TODO: register --tflite, --quantization, --calibration-data, --max-images in the # argparser to enable TFLite export via CLI. Until then, use RFDETR.export(format="tflite"). _ = onnx_path # referenced above; suppress unused-variable warning until CLI is wired up