TimeMarker - Temporal Video Understanding

Powered by TimeMarker-V2-7B on ZeroGPU.

Capabilities:

Temporal Grounding: Find WHEN events happen → [TIME: 45.2s]
Event Counting: Count occurrences → TOTAL: 3
Timestamped Description: Timeline of events
Long Video Support: Handles 1+ hour videos (LVBench Rank #1)

API Endpoints for EagleEye:

POST /call/api_temporal_ground - Find when events happen
POST /call/api_count_events - Count event occurrences
POST /call/api_describe_timeline - Get timestamped description

Upload Video

Query

Task Type

Result

API Usage for EagleEye Integration

Temporal Grounding (Find when events happen)

from gradio_client import Client

client = Client("Cadayn/timemarker-zerogpu")

result = client.predict(
    video_url="https://example.com/match.mp4",
    query="when does the goal happen?",
    num_frames=32,
    api_name="/api_temporal_ground"
)
print(result)
# {"success": True, "events": [{"timestamp_s": 45.2, "description": "Goal scored", ...}], ...}

Event Counting

result = client.predict(
    video_url="https://example.com/match.mp4",
    query="goals scored",
    num_frames=32,
    api_name="/api_count_events"
)
print(result)
# {"success": True, "count": 3, "live_events": [...], "replay_events": [...], ...}

Timestamped Description

result = client.predict(
    video_url="https://example.com/match.mp4",
    focus="player movements and ball possession",
    num_frames=32,
    api_name="/api_describe_timeline"
)
print(result)
# {"success": True, "timeline": [{"start_s": 0.0, "end_s": 10.0, "description": "..."}, ...], ...}