app.py

import os
import json
import random
import shutil
import tempfile
import time
from urllib.parse import urlparse

import cv2
import numpy as np
from PIL import Image
import gradio as gr
from roboflow import Roboflow


def parse_roboflow_url(url: str):
    parsed = urlparse(url)
    parts = parsed.path.strip('/').split('/')
    workspace = parts[0]
    project   = parts[1]
    try:
        version = int(parts[-1])
    except ValueError:
        version = int(parts[-2])
    return workspace, project, version


def convert_seg_to_bbox(api_key: str, dataset_url: str):
    """
    1) Download segmentation dataset from Roboflow
    2) Detect JSON‑vs‑mask export
    3) Convert each mask/polygon to its bounding box (YOLO format)
    4) Preserve original train/valid/test splits
    5) Return before/after visuals + (dataset_path, detection_slug)
    """
    rf = Roboflow(api_key=api_key)
    ws, proj_name, ver = parse_roboflow_url(dataset_url)
    version_obj = rf.workspace(ws).project(proj_name).version(ver)
    dataset     = version_obj.download("coco-segmentation")
    root        = dataset.location

    # scan for any .json files
    all_json = []
    for dp, _, files in os.walk(root):
        for f in files:
            if f.lower().endswith(".json"):
                all_json.append(os.path.join(dp, f))

    if len(all_json) >= 3 and any("train" in os.path.basename(p).lower() for p in all_json):
        # --- COCO‑JSON export branch ---
        # locate train/valid/test JSONs
        json_splits = {}
        for path in all_json:
            fn = os.path.basename(path).lower()
            if "train" in fn:
                json_splits["train"] = path
            elif "val" in fn or "valid" in fn:
                json_splits["valid"] = path
            elif "test" in fn:
                json_splits["test"] = path
        if any(s not in json_splits for s in ("train", "valid", "test")):
            raise RuntimeError(f"Missing one of train/valid/test JSONs: {json_splits}")

        # build category → index from train.json
        train_coco = json.load(open(json_splits["train"], "r"))
        cat_ids    = sorted(c["id"] for c in train_coco.get("categories", []))
        id2idx     = {cid: i for i, cid in enumerate(cat_ids)}

        # aggregate images_info & annotations
        images_info = {}
        annos       = {}
        for split, jf in json_splits.items():
            coco = json.load(open(jf, "r"))
            for img in coco["images"]:
                images_info[img["id"]] = img
            for a in coco["annotations"]:
                xs = a["segmentation"][0][0::2]
                ys = a["segmentation"][0][1::2]
                xmin, xmax = min(xs), max(xs)
                ymin, ymax = min(ys), max(ys)
                w, h       = xmax - xmin, ymax - ymin
                cx, cy     = xmin + w/2, ymin + h/2
                iw = images_info[a["image_id"]]["width"]
                ih = images_info[a["image_id"]]["height"]
                line = (
                    f"{id2idx[a['category_id']]} "
                    f"{cx/iw:.6f} {cy/ih:.6f} {w/iw:.6f} {h/ih:.6f}"
                )
                annos.setdefault(a["image_id"], []).append(line)

        # build filename → path map
        name2id = {img["file_name"]: img["id"] for img in images_info.values()}
        filemap = {
            f: os.path.join(dp, f)
            for dp, _, files in os.walk(root)
            for f in files
            if f in name2id
        }

        # write out per‑split folders
        out_root = tempfile.mkdtemp(prefix="yolov8_")
        for split in ("train", "valid", "test"):
            coco = json.load(open(json_splits[split], "r"))
            img_dir = os.path.join(out_root, split, "images")
            lbl_dir = os.path.join(out_root, split, "labels")
            os.makedirs(img_dir, exist_ok=True)
            os.makedirs(lbl_dir, exist_ok=True)
            for img in coco["images"]:
                fn   = img["file_name"]
                src  = filemap[fn]
                dst  = os.path.join(img_dir, fn)
                txtp = os.path.join(lbl_dir, fn.rsplit(".", 1)[0] + ".txt")
                shutil.copy(src, dst)
                with open(txtp, "w") as f:
                    f.write("\n".join(annos.get(img["id"], [])))

    else:
        # --- Segmentation‐Masks export branch ---
        splits = ["train", "valid", "test"]
        # detect masks subfolder name
        mask_names = ("masks", "mask", "labels")
        out_root   = tempfile.mkdtemp(prefix="yolov8_")

        for split in splits:
            img_dir_src = os.path.join(root, split, "images")
            # find which subdir holds the PNG masks
            mdir = None
            for m in mask_names:
                candidate = os.path.join(root, split, m)
                if os.path.isdir(candidate):
                    mdir = candidate
                    break
            if mdir is None:
                raise RuntimeError(f"No masks folder found under {split}/ (checked {mask_names})")

            img_dir_dst = os.path.join(out_root, split, "images")
            lbl_dir_dst = os.path.join(out_root, split, "labels")
            os.makedirs(img_dir_dst, exist_ok=True)
            os.makedirs(lbl_dir_dst, exist_ok=True)

            for fn in os.listdir(img_dir_src):
                if not fn.lower().endswith((".jpg", ".png")):
                    continue
                src_img  = os.path.join(img_dir_src, fn)
                src_mask = os.path.join(mdir, fn)
                img      = cv2.imread(src_img)
                h, w     = img.shape[:2]

                # read mask & binarize
                mask = cv2.imread(src_mask, cv2.IMREAD_GRAYSCALE)
                _, binm = cv2.threshold(mask, 127, 255, cv2.THRESH_BINARY)
                ys, xs  = np.nonzero(binm)
                if len(xs) == 0:
                    lines = []
                else:
                    xmin, xmax = xs.min(), xs.max()
                    ymin, ymax = ys.min(), ys.max()
                    bw, bh     = xmax - xmin, ymax - ymin
                    cx, cy     = xmin + bw/2, ymin + bh/2
                    # assume single class → index 0
                    lines = [f"0 {cx/w:.6f} {cy/h:.6f} {bw/w:.6f} {bh/h:.6f}"]

                # copy image + write YOLO text
                dst_img = os.path.join(img_dir_dst, fn)
                dst_txt = os.path.join(lbl_dir_dst, fn.rsplit(".",1)[0] + ".txt")
                shutil.copy(src_img, dst_img)
                with open(dst_txt, "w") as f:
                    f.write("\n".join(lines))

    # --- prepare before/after galleries (random sample across out_root) ---
    before, after = [], []
    all_imgs = []
    for split in ("train","valid","test"):
        for fn in os.listdir(os.path.join(out_root, split, "images")):
            path = os.path.join(out_root, split, "images", fn)
            all_imgs.append(path)
    sample = random.sample(all_imgs, min(5, len(all_imgs)))
    for img_path in sample:
        fn       = os.path.basename(img_path)
        img      = cv2.cvtColor(cv2.imread(img_path), cv2.COLOR_BGR2RGB)
        # draw mask outline if available (JSON branch) else read mask again
        seg_vis  = img.copy()
        box_vis  = img.copy()
        # overlay all .txt bboxes
        txtp     = img_path.replace("/images/", "/labels/").rsplit(".",1)[0] + ".txt"
        w, h     = img.shape[1], img.shape[0]
        for line in open(txtp):
            _, cxn, cyn, wnorm, hnorm = map(float, line.split())
            bw, bh = int(wnorm * w), int(hnorm * h)
            x0      = int(cxn * w - bw/2)
            y0      = int(cyn * h - bh/2)
            cv2.rectangle(box_vis, (x0,y0), (x0+bw, y0+bh), (0,255,0), 2)
        before.append(Image.fromarray(seg_vis))
        after.append(Image.fromarray(box_vis))

    detection_slug = proj_name + "-detection"
    return before, after, out_root, detection_slug


def upload_and_train_detection(
    api_key: str,
    detection_slug: str,
    dataset_path: str,
    project_license: str = "MIT",
    project_type: str  = "object-detection"
):
    rf = Roboflow(api_key=api_key)
    ws = rf.workspace()

    # get or create project
    try:
        proj = ws.project(detection_slug)
    except Exception as e:
        if "does not exist" in str(e).lower():
            proj = ws.create_project(
                detection_slug,
                annotation=project_type,
                project_type=project_type,
                project_license=project_license
            )
        else:
            raise

    # upload and kick off train
    _, real_slug = proj.id.rsplit("/", 1)
    ws.upload_dataset(dataset_path, real_slug,
                      project_license=project_license,
                      project_type=project_type)

    try:
        version_num = proj.generate_version(settings={"augmentation":{}, "preprocessing":{}})
    except RuntimeError as e:
        msg = str(e).lower()
        if "unsupported request" in msg or "does not exist" in msg:
            # slug bump fallback
            new_slug = real_slug + "-v2"
            proj = ws.create_project(
                new_slug, annotation=project_type,
                project_type=project_type,
                project_license=project_license
            )
            ws.upload_dataset(dataset_path, new_slug,
                              project_license=project_license,
                              project_type=project_type)
            version_num = proj.generate_version(settings={"augmentation":{}, "preprocessing":{}})
        else:
            raise

    # wait for generation then train
    for _ in range(20):
        try:
            model = proj.version(str(version_num)).train()
            break
        except RuntimeError as e:
            if "still generating" in str(e).lower():
                time.sleep(5)
                continue
            else:
                raise
    else:
        raise RuntimeError("Version generation timed out, try again later.")

    return f"{model['base_url']}{model['id']}?api_key={api_key}"


# --- Gradio UI ---
with gr.Blocks() as app:
    gr.Markdown("## 🔄 Seg→BBox + Auto‑Upload/Train")

    api_input = gr.Textbox(label="Roboflow API Key", type="password")
    url_input = gr.Textbox(label="Segmentation Dataset URL")
    run_btn   = gr.Button("Convert to BBoxes")
    before_g  = gr.Gallery(columns=5, label="Before")
    after_g   = gr.Gallery(columns=5, label="After")
    ds_state  = gr.Textbox(visible=False, label="Dataset Path")
    slug_state= gr.Textbox(visible=False, label="Detection Slug")

    run_btn.click(
        convert_seg_to_bbox,
        inputs=[api_input, url_input],
        outputs=[before_g, after_g, ds_state, slug_state]
    )

    gr.Markdown("## 🚀 Upload & Train Detection Model")
    train_btn = gr.Button("Upload & Train")
    url_out   = gr.Textbox(label="Hosted Model URL")

    train_btn.click(
        upload_and_train_detection,
        inputs=[api_input, slug_state, ds_state],
        outputs=[url_out]
    )

if __name__ == "__main__":
    app.launch()